Suchergebnisse

(This post continues part 1 which just looked at the data. Part 3 on theory is here) When the Fed raises interest rates, how does inflation respond? Are there "long and variable lags" to inflation and output?  There is a standard story: The Fed raises interest rates; inflation is sticky so real interest rates (interest rate - inflation) rise; higher real interest rates lower output and employment; the softer economy pushes inflation down. Each of these is a lagged effect. But despite 40 years of effort, theory struggles to substantiate that story (next post), it's had to see in the data (last post), and the empirical work is ephemeral -- this post.  The vector autoregression and related local projection are today the standard empirical tools to address how monetary policy affects the economy, and have been since Chris Sims' great work in the 1970s. (See Larry Christiano's review.) I am losing faith in the method and results. We need to find new ways to learn about the effects of monetary policy. This post expands on some thoughts on this topic in "Expectations and the Neutrality of Interest Rates," several of my papers from the 1990s* and excellent recent reviews from Valerie Ramey and  Emi Nakamura and Jón Steinsson, who  eloquently summarize the hard identification and computation troubles of contemporary empirical work.Maybe popular wisdom is right, and economics just has to catch up. Perhaps we will. But a popular belief that does not have solid scientific theory and empirical backing, despite a 40 year effort for models and data that will provide the desired answer, must be a bit less trustworthy than one that does have such foundations. Practical people should consider that the Fed may be less powerful than traditionally thought, and that its interest rate policy has different effects than commonly thought. Whether and under what conditions high interest rates lower inflation, whether they do so with long and variable but nonetheless predictable and exploitable lags, is much less certain than you think. Here is a replication of one of the most famous monetary VARs, Christiano Eichenbaum and Evans 1999, from Valerie Ramey's 2016 review: Fig. 1 Christiano et al. (1999) identification. 1965m1–1995m6 full specification: solid black lines; 1983m1–2007m12 full specification: short dashed blue (dark gray in the print version) lines; 1983m1–2007m12, omits money and reserves: long-dashed red (gray in the print version) lines. Light gray bands are 90% confidence bands. Source: Ramey 2016.  Months on x axis. The black lines plot the original specification. The top left panel plots the path of the Federal Funds rate after the Fed unexpectedly raises the interest rate. The funds rate goes up, but only for 6 months or so. Industrial production goes down and unemployment goes up, peaking at month 20. The figure plots the level of the CPI, so inflation is the slope of the lower right hand panel. You see inflation goes the "wrong" way, up, for about 6 months, and then gently declines. Interest rates indeed seem to affect the economy with long lags. This was the broad outline of consensus empirical estimates for many years. It is common to many other studies, and it is consistent with the beliefs of policy makers and analysts. It's pretty much what Friedman (1968) told us to expect. Getting contemporary models to produce something like this is much harder, but that's the next blog post. What's a VAR?I try to keep this blog accessible to nonspecialists, so I'll step back momentarily to explain how we produce graphs like these. Economists who know what a VAR is should skip to the next section heading. How do we measure the effect of monetary policy on other variables? Milton Friedman and Anna Schwartz kicked it off in the Monetary History by pointing to the historical correlation of money growth with inflation and output. They knew as we do that correlation is not causation, so they pointed to the fact that money growth preceeded inflation and output growth.  But as James Tobin pointed out, the cock's crow comes before, but does not cause, the sun to rise. So too people may go get out some money ahead of time when they see more future business activity on the horizon.  Even correlation with a lead is not causation. What to do? Clive Granger's causality and Chris Sims' VAR, especially "Macroeconomics and Reality" gave today's answer. (And there is a reason that everybody mentioned so far has a Nobel prize.) First, we find a monetary policy "shock," a movement in the interest rate (these days; money, then) that is plausibly not a response to economic events and especially to expected future economic events. We think of the Fed setting interest rates by a response to economic data plus deviations from that response, such as interest rate = (#) output + (#) inflation + (#) other variables + disturbance. We want to isolate the "disturbance," movements in the interest rate not taken in response to economic events. (I use "shock" to mean an unpredictable variable, and "disturbance" to mean deviation from an equation like the above, but one that can persist for a while. A monetary policy "shock" is an unexpected movement in the disturbance.)  The "rule" part here can be but need not be the Taylor rule, and can include other variables than output and inflation. It is what the Fed usually does given other variables, and therefore (hopefully) controls for reverse causality from expected future economic events to interest rates. Now, in any individual episode, output and inflation and inflation following a shock will be influenced by subsequent shocks to the economy, monetary and other. But those average out. So, the average value of inflation, output, employment, etc. following a monetary policy shock is a measure of how the shock affects the economy all on its own. That is what has been plotted above. VARs were one of the first big advances in the modern empirical quest to find "exogenous" variation and (somewhat) credibly find causal relationships. Mostly the huge literature varies on how one finds the "shocks." Traditional VARs use regressions of the above equations and the residual is the shock, with a big question just how many and which contemporaneous variables one adds in the regression. Romer and Romer pioneered the "narrative approach," reading the Fed minutes to isolate shocks.  Some technical details at the bottom and much more discussion below. The key is finding shocks. One can just regress output and inflation on the shocks to produce the response function, which is a "local projection" not a "VAR," but I'll use "VAR" for both techniques for lack of a better encompassing word. Losing faithShocks, what shocks?What's a "shock" anyway? The concept is that the Fed considers its forecast of inflation, output and other variables it is trying to control, gauges the usual and appropriate response, and then adds 25 or 50 basis points, at random, just for the heck of it. The question VARS try to answer is the same: What happens to the economy if the Fed raises interest rates unexpectedly, for no particular reason at all? But the Fed never does this. Ask them. Read the minutes. The Fed does not roll dice. They always raise or lower interest rates for a reason, that reason is always a response to something going on in the economy, and most of the time how it affects forecasts of inflation and employment. There are no shocks as defined.I speculated here that we might get around this problem: If we knew the Fed was responding to something that had no correlation with future output, then even though that is an endogenous response, then it is a valid movement for estimating the effect of interest rates on output. My example was, what if the Fed "responds" to the weather. Well, though endogenous, it's still valid for estimating the effect on output. The Fed does respond to lots of things, including foreign exchange, financial stability issues, equity, terrorist attacks, and so forth. But I can't think of any of these in which the Fed is not thinking of these events for their effect on output and inflation, which is why I never took the idea far. Maybe you can. Shock isolation also depends on complete controls for the Fed's information. If the Fed uses any information about future output and inflation that is not captured in our regression, then information about future output and inflation remains in the "shock" series. The famous "price puzzle" is a good example. For the first few decades of VARs, interest rate shocks seemed to lead to higher inflation. It took a long specification search to get rid of this undesired result. The story was, that the Fed saw inflation coming in ways not completely controlled for by the regression. The Fed raised interest rates to try to forestall the inflation, but was a bit hesitant about it so did not cure the inflation that was coming. We see higher interest rates followed by higher inflation, though the true causal effect of interest rates goes the other way. This problem was "cured" by adding commodity prices to the interest rate rule, on the idea that fast-moving commodity prices would capture the information the Fed was using to forecast inflation. (Interestingly these days we seem to see core inflation as the best forecaster, and throw out commodity prices!) With those and some careful orthogonalization choices, the "price puzzle" was tamped down to the one year or so delay you see above. (Neo-Fisherians might object that maybe the price puzzle was trying to tell us something all these years!) Nakamura and Steinsson write of this problem: "What is being assumed is that controlling for a few lags of a few variables captures all endogenous variation in policy... This seems highly unlikely to be true in practice. The Fed bases its policy decisions on a huge amount of data. Different considerations (in some cases highly idiosyncratic) affect policy at different times. These include stress in the banking system, sharp changes in commodity prices, a recent stock market crash, a financial crisis in emerging markets, terrorist attacks, temporary investment tax credits, and the Y2K computer glitch. The list goes on and on. Each of these considerations may only affect policy in a meaningful way on a small number of dates, and the number of such influences is so large that it is not feasible to include them all in a regression. But leaving any one of them out will result in a monetary policy "shock" that the researcher views as exogenous but is in fact endogenous." Nakamura and Steinsson offer 9/11 as another example summarizing my "high frequency identification" paper with Monika Piazzesi: The Fed lowered interest rates after the terrorist attack, likely reacting to its consequences for output and inflation. But VARs register the event as an exogenous shock.Romer and Romer suggested that we use Fed Greenbook forecasts of inflation and output as controls, as those should represent the Fed's complete information set. They provide narrative evidence that Fed members trust Greenback forecasts more than you might suspect. This issue is a general Achilles heel of empirical macro and finance: Does your procedure assume agents see no more information than you have included in the model or estimate? If yes, you have a problem. Similarly, "Granger causality" answers the cock's crow-sunrise problem by saying that if unexpected x leads unexpected y then x causes y. But it's only real causality if the "expected" includes all information, as the price puzzle counterexample shows. Just what properties do we need of a shock in order to measure the response to the question, "what if the Fed raised rates for no reason?" This strikes me as a bit of an unsolved question -- or rather, one that everyone thinks is so obvious that we don't really look at it. My suggestion that the shock only need be orthogonal to the variable whose response we're estimating is informal, and I don't know of formal literature that's picked it up. Must "shocks" be unexpected, i.e. not forecastable from anything in the previous time information set? Must they surprise people? I don't think so -- it is neither necessary nor sufficient for  shock to be unforecastable for it to identify the inflation and output responses. Not responding to expected values of the variable whose response you want to measure should be enough.  If bond markets found out about a random funds rate rise one day ahead, it would then be an "expected" shock, but clearly just as good for macro. Romer and Romer have been criticized that their shocks are predictable, but this may not matter. The above Nakamura and Steinsson quote says leaving out any information leads to a shock that is not strictly exogenous. But strictly exogenous may not be necessary for estimating, say, the effect of interest rates on inflation. It is enough to rule out reverse causality and third effects. Either I'm missing a well known econometric literature, as is everyone else writing the VARs I've read who don't cite it, or there is a good theory paper to be written.Romer and Romer, thinking deeply about how to read "shocks" from the Fed minutes, define shocks thus to circumvent the "there are no shocks" problem:we look for times when monetary policymakers felt the economy was roughly at potential (or normal) output, but decided that the prevailing rate of inflation was too high. Policymakers then chose to cut money growth and raise interest rates, realizing that there would be (or at least could be) substantial negative consequences for aggregate output and unemployment. These criteria are designed to pick out times when policymakers essentially changed their tastes about the acceptable level of inflation. They weren't just responding to anticipated movements in the real economy and inflation. [My emphasis.]  You can see the issue. This is not an "exogenous" movement in the funds rate. It is a response to inflation, and to expected inflation, with a clear eye on expected output as well. It really is a nonlinear rule, ignore inflation for a while until it gets really bad then finally get serious about it. Or, as they say, it is a change in rule, an increase in the sensitivity of the short run interest rate response to inflation, taken in response to inflation seeming to get out of control in a longer run sense. Does this identify the response to an "exogenous" interest rate increase? Not really.  But maybe it doesn't matter. Are we even asking an interesting question? The whole question, what would happen if the Fed raised interest rates for no reason, is arguably besides the point. At a minimum, we should be clearer about what question we are asking, and whether the policies we analyze are implementations of that question. The question presumes a stable "rule," (e.g. \(i_t = \rho i_{t-1} + \phi_\pi \pi_t + \phi_x x_t + u_t\)) and asks what happens in response to a deviation \( +u_t \) from the rule. Is that an interesting question? The standard story for 1980-1982 is exactly not such an event. Inflation was not conquered by a big "shock," a big deviation from 1970s practice, while keeping that practice intact. Inflation was conquered (so the story goes) by a change in the rule, by a big increase in $\phi_\pi$. That change raised interest rates,  but arguably without any deviation from the new rule \(u_t\) at all. Thinking in terms of the Phillips curve \( \pi_t = E_t \pi_{t+1} + \kappa x_t\), it was not a big negative \(x_t\) that brought down inflation, but the credibility of the new rule that brought down \(E_t \pi_{t+1}\). If the art of reducing inflation is to convince people that a new regime has arrived, then the response to any monetary policy "shock" orthogonal to a stable "rule" completely misses that policy. Romer and Romer are almost talking about a rule-change event. For 2022, they might be looking at the Fed's abandonment of flexible average inflation targeting and its return to a Taylor rule. However, they don't recognize the importance of the distinction, treating changes in rule as equivalent to a residual. Changing the rule changes expectations in quite different ways from a residual of a stable rule. Changes with a bigger commitment should have bigger effects, and one should standardize somehow by the size and permanence of the rule change, not necessarily the size of the interest rate rise. And, having asked "what if the Fed changes rule to be more serious about inflation," we really cannot use the analysis to estimate what happens if the Fed shocks interest rates and does not change the rule. It takes some mighty invariance result from an economic theory that a change in rule has the same effect as a shock to a given rule. There is no right and wrong, really. We just need to be more careful about what question the empirical procedure asks, if we want to ask that question, and if our policy analysis actually asks the same question. Estimating rules, Clarida Galí and Gertler. Clarida, Galí, and Gertler (2000) is a justly famous paper, and in this context for doing something totally different to evaluate monetary policy. They estimate rules, fancy versions of \(i_t = \rho i_{t-1} +\phi_\pi \pi_t + \phi_x x_t + u_t\), and they estimate how the \(\phi\) parameters change over time. They attribute the end of 1970s inflation to a change in the rule, a rise in \(\phi_\pi\) from the 1970s to the 1980s. In their model, a higher \( \phi_\pi\) results in less volatile inflation. They do not estimate any response functions. The rest of us were watching the wrong thing all along. Responses to shocks weren't the interesting quantity. Changes in the rule were the interesting quantity. Yes, I criticized the paper, but for issues that are irrelevant here. (In the new Keynesian model, the parameter that reduces inflation isn't the one they estimate.) The important point here is that they are doing something completely different, and offer us a roadmap for how else we might evaluate monetary policy if not by impulse-response functions to monetary policy shocks. Fiscal theoryThe interesting question for fiscal theory is, "What is the effect of an interest rate rise not accompanied by a change in fiscal policy?" What can the Fed do by itself? By contrast, standard models (both new and old Keynesian) include concurrent fiscal policy changes when interest rates rise. Governments tighten in present value terms, at least to pay higher interest costs on the debt and the windfall to bondholders that flows from unexpected disinflation. Experience and estimates surely include fiscal changes along with monetary tightening. Both fiscal and monetary authorities react to inflation with policy actions and reforms. Growth-oriented microeconomic reforms with fiscal consequences often follow as well -- rampant inflation may have had something to do with Carter era trucking, airline, and telecommunications reform. Yet no current estimate tries to look for a monetary shock orthogonal to fiscal policy change. The estimates we have are at best the effects of monetary policy together with whatever induced or coincident fiscal and microeconomic policy tends to happen at the same time as central banks get serious about fighting inflation. Identifying the component of a monetary policy shock orthogonal to fiscal policy, and measuring its effects is  a first order question for fiscal theory of monetary policy. That's why I wrote this blog post. I set out to do it, and then started to confront how VARs are already falling apart in our hands. Just what "no change in fiscal policy" means is an important question that varies by application. (Lots more in "fiscal roots" here, fiscal theory of monetary policy here and in FTPL.) For simple calculations, I just ask what happens if interest rates change with no change in primary surplus. One might also define "no change" as no change in tax rates, automatic stabilizers, or even habitual discretionary stimulus and bailout, no disturbance \(u_t\) in a fiscal rule \(s_t = a + \theta_\pi \pi_t + \theta_x x_t + ... + u_t\). There is no right and wrong here either, there is just making sure you ask an interesting question. Long and variable lags, and persistent interest rate movementsThe first plot shows a mighty long lag between the monitor policy shock and its effect on inflation and output. That does not mean that the economy has long and variable lags. This plot is actually not representative, because in the black lines the interest rate itself quickly reverts to zero. It is common to find a more protracted interest rate response to the shock, as shown in the red and blue lines. That mirrors common sense: When the Fed starts tightening, it sets off a year or so of stair-step further increases, and then a plateau, before similar stair-step reversion. That raises the question, does the long-delayed response of output and inflation represent a delayed response to the initial monetary policy shock, or does it represent a nearly instantaneous response to the higher subsequent interest rates that the shock sets off? Another way of putting the question, is the response of inflation and output invariant to changes in the response of the funds rate itself? Do persistent and transitory funds rate changes have the same responses? If you think of the inflation and output responses as economic responses to the initial shock only, then it does not matter if interest rates revert immediately to zero, or go on a 10 year binge following the initial shock. That seems like a pretty strong assumption. If you think that a more persistent interest rate response would lead to a larger or more persistent output and inflation response, then you think some of what we see in the VARs is a quick structural response to the later higher interest rates, when they come. Back in 1988, I posed this question in "what do the VARs mean?" and showed you can read it either way. The persistent output and inflation response can represent either long economic lags to the initial shock, or much less laggy responses to interest rates when they come. I showed how to deconvolute the response function to the structural effect of interest rates on inflation and output and how persistently interest rates rise.  The inflation and output responses might be the same with shorter funds rate responses, or they might be much different. Obviously (though often forgotten), whether the inflation and output responses are invariant to changes in the funds rate response needs a model. If in the economic model only unexpected interest rate movements affect output and inflation, though with lags, then the responses are as conventionally read structural responses and invariant to the interest rate path. There is no such economic model. Lucas (1972) says only unexpected money affects output, but with no lags, and expected money affects inflation. New Keynesian models have very different responses to permanent vs. transitory interest rate shocks. Interestingly, Romer and Romer do not see it this way, and regard their responses as structural long and variable lags, invariant to the interest rate response. They opine that given their reading of a positive shock in 2022, a long and variable lag to inflation reduction is baked in, no matter what the Fed does next. They argue that the Fed should stop raising interest rates. (In fairness, it doesn't look like they thought about the issue much, so this is an implicit rather than explicit assumption.) The alternative view is that effects of a shock on inflation are really effects of the subsequent rate rises on inflation, that the impulse response function to inflation is not invariant to the funds rate response, so stopping the standard tightening cycle would undo the inflation response. Argue either way, but at least recognize the important assumption behind the conclusions. Was the success of inflation reduction in the early 1980s just a long delayed response to the first few shocks? Or was the early 1980s the result of persistent large real interest rates following the initial shock? (Or, something else entirely, a coordinated fiscal-monetary reform... But I'm staying away from that and just discussing conventional narratives, not necessarily the right answer.) If the latter, which is the conventional narrative, then you think it does matter if the funds rate shock is followed by more funds rate rises (or positive deviations from a rule), that the output and inflation response functions do not directly measure long lags from the initial shock. De-convoluting the structural funds rate to inflation response and the persistent funds rate response, you would estimate much shorter structural lags. Nakamura and Steinsson are of this view: While the Volcker episode is consistent with a large amount of monetary nonneutrality, it seems less consistent with the commonly held view that monetary policy affects output with "long and variable lags." To the contrary, what makes the Volcker episode potentially compelling is that output fell and rose largely in sync with the actions [interest rates, not shocks] of the Fed. And that's a good thing too. We've done a lot of dynamic economics since Friedman's 1968 address. There is really nothing in dynamic economic theory that produces a structural long-delayed response to shocks, without the continued pressure of high interest rates. (A correspondent objects to "largely in sync" pointing out several clear months long lags between policy actions and results in 1980. It's here for the methodological point, not the historical one.) However, if the output and inflation responses are not invariant to the interest rate response, then the VAR directly measures an incredibly narrow experiment: What happens in response to a surprise interest rate rise, followed by the plotted path of interest rates? And that plotted path is usually pretty temporary, as in the above graph. What would happen if the Fed raised rates and kept them up, a la 1980? The VAR is silent on that question. You need to calibrate some model to the responses we have to infer that answer. VARs and shock responses are often misread as generic theory-free estimates of "the effects of monetary policy." They are not. At best, they tell you the effect of one specific experiment: A random increase in funds rate, on top of a stable rule, followed by the usual following path of funds rate. Any other implication requires a model, explicit or implicit. More specifically, without that clearly false invariance assumption, VARs cannot directly answer a host of important questions. Two on my mind: 1) What happens if the Fed raises interest rates permanently? Does inflation eventually rise? Does it rise in the short run? This is the "Fisherian" and "neo-Fisherian" questions, and the answer "yes" pops unexpectedly out of the standard new-Keynesian model. 2) Is the short-run negative response of inflation to interest rates stronger for more persistent rate rises? The long-term debt fiscal theory mechanism for a short-term inflation decline is tied to the persistence of the shock and the maturity structure of the debt. The responses to short-lived interest rate movements (top left panel) are silent on these questions. Directly is an important qualifier. It is not impossible to answer these questions, but you have to work harder to identify persistent interest rate shocks. For example, Martín Uribe identifies permanent vs. transitory interest rate shocks, and finds a positive response of inflation to permanent interest rate rises. How? You can't just pick out the interest rate rises that turned out to be permanent. You have to find shocks or components of the shock that are ex-ante predictably going to be permanent, based on other forecasting variables and the correlation of the shock with other shocks. For example, a short-term rate shock that also moves long-term rates might be more permanent than one which does not do so. (That requires the expectations hypothesis, which doesn't work, and long term interest rates move too much anyway in response to transitory funds rate shocks. So, this is not directly a suggestion, just an example of the kind of thing one must do. Uribe's model is more complex than I can summarize in a blog.)  Given how small and ephemeral the shocks are already, subdividing them into those that are expected to have permanent vs. transitory effects on the federal funds rate is obviously a challenge. But it's not impossible. Monetary policy shocks account for small fractions of inflation, output and funds rate variation. Friedman thought that most recessions and inflations were due to monetary mistakes. The VARs pretty uniformly deny that result. The effects of monetary policy shocks on output and inflation add up to less than 10 percent of the variation of output and inflation. In part the shocks are small, and in part the responses to the shocks are small. Most recessions come from other shocks, not monetary mistakes.  Worse, both in data and in models, most inflation variation comes from inflation shocks, most output variation comes from output shocks, etc. The cross-effects of one variable on another are small. And "inflation shock" (or "marginal cost shock"),  "output shock" and so forth are just labels for our ignorance -- error terms in regressions, unforecasted movements -- not independently measured quantities. (This and old point, for example in my 1994 paper with the great title "Shocks." Technically, the variance of output is the sum of the squares of the impulse-response functions -- the plots -- times the variance of the shocks. Thus small shocks and small responses mean not much variance explained.)This is a deep point. The exquisite attention put to the effects of monetary policy in new-Keynesian models, while interesting to the Fed, are then largely beside the point if your question is what causes recessions. Comprehensive models work hard to match all of the responses, not just to monetary policy shocks. But it's not clear that the nominal rigidities that are important for the effects of monetary policy are deeply important to other (supply) shocks, and vice versa. This is not a criticism. Economics always works better if we can use small models that focus on one thing -- growth, recessions, distorting effect of taxes, effect of monetary policy -- without having to have a model of everything in which all effects interact. But, be clear we no longer have a model of everything.  "Explaining recessions" and "understanding the effects of monetary policy" are somewhat separate questions. Monetary policy shocks also account for small fractions of the movement in the federal funds rate itself. Most of the funds rate movement is in the rule, the reaction to the economy term. Like much empirical economics, the quest for causal identification leads us to look at a tiny causes with tiny effects, that do little to explain much variation in the variable of interest (inflation). Well, cause is cause, and the needle is the sharpest item in the haystack. But one worries about the robustness of such tiny effects, and to what extent they summarize historical experience.  To be concrete, here is a typical shock regression, 1960:1-2023:6 monthly data, standard errors in parentheses: ff(t) = a + b ff(t-1) + c[ff(t-1)-ff(t-2)] + d CPI(t) + e unemployment(t) + monetary policy shock, Where "CPI" is the percent change in the CPI (CPIAUCSL) from a year earlier. ff(t-1)ff(t-1)-ff(t-2)CPIUnempR20.970.390.032-0.0170.985(0.009)(0.07)(0.013)(0.009)The funds rate is persistent -- the lag term (0.97) is large. Recent changes matter too: Once the Fed starts a tightening cycle, it's likely to keep raising rates. And the Fed responds to CPI and unemployment. The plot shows the actual federal funds rate (blue), the model or predicted federal funds rate (red), the shock which is the difference between the two (orange) and the Romer and Romer dates (vertical lines). You can't see the difference between actual and predicted funds rate, which is the point. They are very similar and the shocks are small. They are closer horizontally than vertically, so the vertical difference plotted as shock is still visible. The shocks are much smaller than the funds rate, and smaller than the rise and fall in the funds rate in a typical tightening or loosening cycle. The shocks are bunched, with by far the biggest ones in the early 1980s. The shocks have been tiny since the 1980s. (Romer and Romer don't find any shocks!) Now, our estimates of the effect of monetary policy look at the average values of inflation, output, and employment in the 4-5 years after a shock. Really, you say, looking at the graph? That's going to be dominated by the experience of the early 1980s. And with so many positive and negative shocks close together, the average value 4 years later is going to be driven by subtle timing of when the positive or negative shocks line up with later events. Put another way, here is a plot of inflation 30 months after a shock regressed on the shock. Shock on the x axis, subsequent inflation on the y axis. The slope of the line  is our estimate of the effect of the shock on inflation 30 months out (source, with details). Hmm. One more graph (I'm having fun here):This is a plot of inflation for the 4 years after each shock, times that shock. The right hand side is the same graph with an expanded y scale. The average of these histories is our impulse response function.  (The big lines are the episodes which multiply the big shocks of the early 1980s. They mostly converge because, either multiplied by positive or negative shocks, inflation wend down in the 1980s.) Impulse response functions are just quantitative summaries of the lessons of history. You may be  underwhelmed that history is sending a clear story.  Again, welcome to causal economics -- tiny average responses to tiny but identified movements is what we estimate, not broad lessons of history. We do not estimate "what is the effect of the sustained high real interest rates of the early 1980s," for example, or "what accounts for the sharp decline of inflation in the early 1980s?" Perhaps we should, though confronting endogeneity of the interest rate responses some other way. That's my main point today. Estimates disappear after 1982Ramey's first variation in the first plot is to use data from 1983 to 2007. Her second variation is to also omit the monetary variables. Christiano Eichenbaum and Evans were still thinking in terms of money supply control, but our Fed does not control money supply. The evidence that higher interest rates lower inflation disappears after 1983, with or without money. This too is a common finding. It might be because there simply aren't any monetary policy shocks. Still, we're driving a car with a yellowed AAA road map dated 1982 on it.  Monetary policy shocks still seem to affect output and employment, just not inflation. That poses a deeper problem. If there just aren't any monetary policy shocks, we would just get big standard errors on everything. That only inflation disappears points to the vanishing Phillips curve, which will be the weak point in the theory to come. It is the Phillips curve by which lower output and employment push down inflation. But without the Phillips curve, the whole standard story for interest rates to affect inflation goes away. Computing long-run responsesThe long lags of the above plot are already pretty long horizons, with interesting economics still going on at 48 months. As we get interested in long run neutrality, identification via long run sign restrictions (monetary policy should not permanently affect output), and the effect of persistent interest rate shocks, we are interested in even longer run responses. The "long run risks" literature in asset pricing is similarly crucially interested in long run properties. Intuitively, we should know this will be troublesome. There aren't all that many nonoverlapping 4 year periods after interest rate shocks to measure effects, let alone 10 year periods.VARs estimate long run responses with a parametric structure. Organize the data (output, inflation, interest rate, etc) into a vector \(x_t = [y_t \; \pi_t \; i_t \; ...]'\), then the VAR can be written \(x_{t+1} = Ax_t + u_t\). We start from zero, move \(x_1 = u_1\) in an interesting way, and then the response function just simulates forward, with \(x_j = A^j x_1\).  But here an oft-forgotten lesson of 1980s econometrics pops up: It is dangerous to estimate long-run dynamics by fitting a short run model and then finding its long-run implications. Raising matrices to the 48th power \(A^{48}\) can do weird things, the 120th power (10 years) weirder things. OLS and maximum likelihood prize one step ahead \(R^2\), and will happily accept small one step ahead mis specifications that add up to big misspecification 10 years out. (I learned this lesson in the "Random walk in GNP.") Long run implications are driven by the maximum eigenvalue of the \(A\) transition matrix, and its associated eigenvector. \(A^j = Q \Lambda^j Q^{-1}\).  This is a benefit and a danger. Specify and estimate the dynamics of the combination of variables with the largest eigenvector  right, and lots of details can be wrong. But standard estimates aren't trying hard to get these right. The "local projection" alternative directly estimates long run responses: Run regressions of inflation in 10 years on the shock today. You can see the tradeoff: there aren't many non-overlapping 10 year intervals, so this will be imprecisely estimated. The VAR makes a strong parametric assumption about long-run dynamics. When it's right, you get better estimates. When it's wrong, you get misspecification. My experience running lots of VARs is that monthly VARs raised to large powers often give unreliable responses. Run at least a one-year VAR before you start looking at long run responses. Cointegrating vectors are the most reliable variables to include. They are typically the state variable that most reliably carries long - run responses. But pay attention to getting them right. Imposing integrating and cointegrating structure by just looking at units is a good idea. The regression of long-run returns on dividend yields is a good example. The dividend yield is a cointegrating vector, and is the slow-moving state variable. A one period VAR \[\left[ \begin{array}{c} r_{t+1} \\ dp_{t+1} \end{array} \right] = \left[ \begin{array}{cc} 0 & b_r  \\ 0 & \rho \end{array}\right]    \left[ \begin{array}{c} r_{t} \\ dp_{t} \end{array}\right]+ \varepsilon_{t+1}\] implies a long horizon regression \(r_{t+j} = b_r \rho^j dp_{t} +\) error. Direct regressions ("local projections") \(r_{t+j} = b_{r,j} dp_t + \) error give about the same answers, though the downward bias in \(\rho\) estimates is a bit of an issue, but with much larger standard errors. The constraint \(b_{r,j} = b_r \rho^j\) isn't bad. But it can easily go wrong. If you don't impose that dividends and price are cointegrated, or with vector other than 1 -1, if you  allow a small sample to estimate \(\rho>1\), if you don't put in dividend yields at all and just a lot of short-run forecasters, it can all go badly. Forecasting bond returns was for me a good counterexample. A VAR forecasting one-year bond returns  from today's yields gives very different results from taking a monthly VAR, even with several lags, and using \(A^{12}\) to infer the one-year return forecast. Small pricing errors or microstructure dominate the monthly data, which produces junk when raised to the twelfth power. (Climate regressions are having fun with the same issue. Small estimated effects of temperature on growth, raised to the 100th power, can produce nicely calamitous results. But use basic theory to think about units.)  Nakamura and Steinsson (appendix) show how sensitive some standard estimates of impulse response functions are to these questions. Weak evidenceFor the current policy question, I hope you get a sense of how weak the evidence is for the "standard view" that higher interest rates reliably lower inflation, though with a long and variable lag, and the Fed has a good deal of control over inflation. Yes, many estimates look the same, but there is a pretty strong prior going in to that. Most people don't publish papers that don't conform to something like the standard view. Look how long it took from Sims (1980) to Christiano Eichenbaum and Evans (1999) to produce a response function that does conform to the standard view, what Friedman told us to expect in (1968). That took a lot of playing with different orthogonalization, variable inclusion, and other specification assumptions. This is not criticism: when you have a strong prior, it makes sense to see if the data can be squeezed in to the prior. Once authors like Ramey and Nakamura and Steinsson started to look with a critical eye, it became clearer just how weak the evidence is. Standard errors are also wide, but the variability in results due to changes in sample and specification are much larger than formal standard errors. That's why I don't stress that statistical aspect. You play with 100 models, try one variable after another to tamp down the price puzzle, and then compute standard errors as if the 100th model were written in stone. This post is already too long, but showing how results change with different specifications would have been a good addition. For example, here are a few more Ramey plots of inflation responses, replicating various previous estimatesTake your pick. What should we do instead? Well, how else should we measure the effects of monetary policy? One natural approach turns to the analysis of historical episodes and changes in regime, with specific models in mind. Romer and Romer pass on thoughts on this approach: ...some macroeconomic behavior may be fundamentally episodic in nature. Financial crises, recessions, disinflations, are all events that seem to play out in an identifiable pattern. There may be long periods where things are basically fine, that are then interrupted by short periods when they are not. If this is true, the best way to understand them may be to focus on episodes—not a cross-section proxy or a tiny sub-period. In addition, it is valuable to know when the episodes were and what happened during them. And, the identification and understanding of episodes may require using sources other than conventional data.A lot of my and others'  fiscal theory writing has taken a similar view. The long quiet zero bound is a test of theories: old-Keynesian models predict a delation spiral, new-Keynesian models predicts sunspot volatility, fiscal theory is consistent with stable quiet inflation.  The emergence of inflation in 2021 and its easing despite interest rates below inflation likewise validates fiscal vs. standard theories.  The fiscal implications of abandoning the gold standard in 1933 plus Roosevelt's "emergency" budget make sense of that episode. The new-Keynesian reaction parameter \(\phi_\pi\) in \(i_t - \phi_\pi \pi_t\), which leads to unstable dynamics for ](\phi_\pi>1\) is not identified by time series data. So use "other sources," like plain statements on the Fed website about how they react to inflation. I already cited Clarida Galí and Gertler, for measuring the rule not the response to the shock, and explaining the implications of that rule for their model. Nakamura and Steinsson likewise summarize Mussa's (1986) classic study of what happens when countries switch from fixed to floating exchange rates: "The switch from a fixed to a flexible exchange rate is a purely monetary action. In a world where monetary policy has no real effects, such a policy change would not affect real variables like the real exchange rate. Figure 3 demonstrates dramatically that the world we live in is not such a world."Also, analysis of particular historical episodes is enlightening. But each episode has other things going on and so invites alternative explanations. 90 years later, we're still fighting about what caused the Great Depression. 1980 is the poster child for monetary disinflation, yet as Nakamura and Steinsson write, Many economists find the narrative account above and the accompanying evidence about output to be compelling evidence of large monetary nonneutrality. However, there are other possible explanations for these movements in output. There were oil shocks both in September 1979 and in February 1981.... Credit controls were instituted between March and July of 1980. Anticipation effects associated with the phased-in tax cuts of the Reagan administration may also have played a role in the 1981–1982 recession ....Studying changes in regime, such as fixed to floating or the zero bound era, help somewhat relative to studying a particular episode, in that they have some of the averaging of other shocks. But the attraction of VARs will remain. None of these produces what VARs seemed to produce, a theory-free qualitative estimate of the effects of monetary policy. Many tell you that prices are sticky, but not how prices are sticky. Are they old-Keynesian backward looking sticky or new-Keynesian rational expectations sticky? What is the dynamic response of relative inflation to a change in a pegged exchange rate? What is the dynamic response of real relative prices to productivity shocks? Observations such as Mussa's graph can help to calibrate models, but does not answer those questions directly.  My observations about the zero bound or the recent inflation similarly seem (to me) decisive about one class of model vs. another, at least subject to Occam's razor about epicycles, but likewise do not provide a theory-free  impulse response function.  Nakamura and Steinsson write at length about other approaches; model-based moment matching and use of micro data in particular. This post is going on too long; read their paper. Of course, as we have seen, VARs only seem to offer a model-free quantitative measurement of "the effects of monetary policy," but it's hard to give up on the appearance of such an answer. VARs and impulse responses also  remain very useful ways of summarizing the correlations and cross correlations of data, even without cause and effect interpretation. In the end, many ideas are successful in economics when they tell researchers what to do, when they offer a relatively clear recipe for writing papers. "Look at episodes and think hard is not such  recipe." "Run a VAR is." So, as you think about how we can evaluate monetary policy, think about a better recipe as well as a good answer.  (Stay tuned. This post is likely to be updated a few times!) VAR technical appendixTechnically, running VARs is very easy, at least until you start trying to smooth out responses with Bayesian  and other techniques. Line up the data in a vector, i.e. \(x_t = [i_t \; \pi_t\; y_t]'\). Then run a regression of each variable on lags of the others, \[x_t = Ax_{t-1} + u_t.\] If you want more than one lag of the right hand variables, just make a bigger \(x\) vector,  \(x_t = [i_t\; \pi_t \; y_t \; i_{t-1}\; \pi_{t-1} \;y_{t-1}]'.\) The residuals of such regressions \(u_t\) will be correlated, so you have to decide whether, say, the correlation between interest rate and inflation shocks means the Fed responds in the period to inflation, or inflation responds within the period to interest rates, or some combination of the two. That's the  "identification" assumption issue. You can write it as a matrix \(C\) so that \(u_t = C \varepsilon_t\) and cov\((\varepsilon_t \varepsilon_t')=I\) or you can include some contemporaneous values into the right hand sides. Now, with \(x_t = Ax_{t-1} + C\varepsilon_t\), you start with \(x_0=0\), choose one series to shock, e.g. \(\varepsilon_{i,1}=1\) leaving the others alone, and just simulate forward. The resulting path of the other variables is the above plot, the "impulse response function." Alternatively you can run a regression \(x_t = \sum_{j=0}^\infty \theta_j \varepsilon_{t-j}\) and the \(\theta_j\) are (different, in sample) estimates of the same thing. That's "local projection". Since the right hand variables are all orthogonal, you can run single or multiple regressions. (See here for equations.) Either way, you have found the moving average representation, \(x_t = \theta(L)\varepsilon_t\), in the first case with \(\theta(L)=(I-AL)^{-1}C\) in the second case directly. Since the right hand variables are all orthogonal, the variance of the series is the sum of its loading on all of the shocks, \(cov(x_t) = \sum_{j=0}^\infty \theta_j \theta_j'\). This "forecast error variance decomposition" is behind my statement that small amounts of inflation variance are due to monetary policy shocks rather than shocks to other variables, and mostly inflation shocks.   Update:Luis Garicano has a great tweet thread explaining the ideas with a medical analogy. Kamil Kovar has a nice follow up blog post, with emphasis on Europe. He makes a good point that I should have thought of: A monetary policy "shock" is a deviation from a "rule." So, the Fed's and ECB's failure to respond to inflation as they "usually" do in 2021-2022 counts exactly the same as a 3-5% deliberate lowering of the interest rate. Lowering interest rates for no reason, and leaving interest rates alone when the regression rule says raise rates are the same in this methodology.  That "loosening" of policy was quickly followed by inflation easing, so an updated VAR should exhibit a strong "price puzzle" -- a negative shock is followed by less, not more inflation. Of course historians and practical people might object that failure to act as usual has exactly the same effects as acting. * Some Papers: Comment on Romer and Romer What ends recessions? Some "what's a shock?"Comment on Romer and Romer A new measure of monetary policy. The greenbook forecasts, and beginning thoughts that strict exogeneity is not necessary. Shocks monetary shocks explain small fractions of output variance.Comments on Hamilton, more thoughts on what a shock is.What do the VARs mean? cited above, is the response to the shock or to persistent interest rates?The Fed and Interest Rates, with Monika Piazzesi. Daily data and interest rates to identify shocks. Decomposing the yield curve with Monika Piazzesi. Starts with a  great example of how small changes in specification lead to big differences in long run forecasts. Time seriesA critique of the application of unit root tests pretesting for unit roots and cointegration is a bad ideaHow big is the random walk in GNP? lessons in not using short run dynamics to infer long run properties. Permanent and transitory components of GNP and stock prices a favorite of cointegration really helps on long run propertiesTime series for macroeconomics and finance notes that never quite became a book. Explains VARs and responses. 

Does education matter for growth? Which type of education investment (primary, secondary, or tertiary) matters most? Is there a relationship between growth or innovation and the governance of higher education? This paper surveys recent attempts at answering these questions. It first contrasts the 'Lucas approach,' whereby growth is affected by the accumulation of human capital, with the 'Nelson?Phelps approach,' whereby growth is affected by the stock of human capital and by its interaction with the underlying process of technological innovation. Then the paper argues that growth in countries that are close to the world technological frontier benefit more from tertiary education, whereas countries that lie below the frontier benefit more from primary and secondary education. Finally, the paper discusses the relationship between innovation and the governance of universities.

This guide helps researchers in all fields share their findings, knowledge, and ideas effectively and beyond typical silos. By pursuing the recommendations in this book, researchers can increase the exposure of their work to increase impact. Chapters cover the most effective ways to share readers' research, such as: social media (maximizing exposure and branding, leveraging time-saving tools) media (landing interviews and contributing to public dialogue) writing (landing book deals and succeeding in key writing opportunities) speaking (giving TED Talks, delivering conference keynote presentations, appearing on NPR, etc.) connecting (networking, influencing policy, etc.) honors (winning awards and recognition to expand your platform) Rich in tips, strategies, and guidelines, this book also includes clever "fast tracks" and downloadable online resources that provide links, leads, and templates to help secure radio broadcasts, podcasts, publications, conferences, awards, and other opportunities. {"references": ["Rankin, J. G. (2020). Increasing the impact of your research: A practical guide to sharing your findings and widening your reach. New York, NY: Routledge/Taylor & Francis."]}

Empirical work in labor economics has focused on rent sharing as an explanation for the observed correlation between wages and profitability. The alternative explanation of risk sharing between workers and employers has not been tested. Using a unique panel data set for four African countries, Authors find strong evidence of risk sharing. Workers in effect offer insurance to employers: when firms are hit by temporary shocks, the effect on profits is cushioned by risk sharing with workers. Rent sharing is a symptom of an inefficient labor market. Risk sharing; by contrast, can be seen as an efficient response to missing markets. Authors evidence suggests that risk sharing accounts for a substantial part of the observed effect of shocks on wages.

Entender el territorio a partir del funcionamiento de la economía y, en forma particular, a partir del comportamiento espacial del empleo. A través de la formación de áreas geográficas para el reporte de estadísticas laborales. Midiendo y comparando la movilidad y los procesos de articulación y expansión de mercados laborales locales (MLL). Cuantificando el impacto que tiene el capital humano y la estructura laboral en la variación del empleo a nivel regional. Y ubicando geográficamente Distritos Industriales. En el primer ensayo titulado: Áreas de viaje al trabajo en España un análisis estadístico mutivariante. Se investigó como formar grupos de unidades geográficas determinadas a partir de los vínculos laborales compartidos entre municipios. La muestra fue dividida en nueve estratos. Para cada estrato se capturó la vinculación laboral entre municipios en tablas de contingencia bidimensional que muestran la cantidad los viajes al trabajo y los orígenes y destinos de las personas ocupadas en cada municipio del país. Partiendo de aquí se aplicó Análisis de Conglomerados Jerárquicos para conformar Mercados Laborales Locales (MLL). Posteriormente con el fin de medir y comparar la movilidad y los procesos de articulación y expansión de cada MLL, se analizó el comportamiento diferencial entre los años 2001 y 2011 de cuatro variables en cada MLL obtenido: autonomía de la oferta, autonomía de la demanda, empleo y cantidad de trabajadores. Teniendo en cuenta la evolución entre 2001 y 2011 de estas cuatro variables se clasificó cada MLL en cuatro comportamientos diferentes. También se encontró que el la situación de mercados superavitarios en empleos en el 2001 de los estratos hombres, mujeres y trabajadores manuales calificados, paso a ser deficitario en el 2011. El único estrato que mostro un comportamiento superavitario fue el de agricultura. Segundo ensayo titulado: Evaluación del desarrollo económico medido con variables asociadas a educación en zonas comerciales derivadas de economías de aglomeración y de ubicación en España. En este ensayo se modeló con el enfoque de regresión geográficamente ponderada, la variación del empleo en España entre 2001 y 2011 en función del capital humano y la estructura laboral que habitaba en cada municipio de la muestra en el 2001. Al utilizar regresión geográficamente ponderada se obtuvo la respuesta a las influencias geográficamente próximas de los municipios circundantes, dicha información fue resumida en mapas. Con el fin de conocer el comportamiento del empleo en función del capital humano y estructura laboral a través de todo el país. Indicando aquellas zonas con mayores oportunidades de negocio potencial. La heterogeneidad espacial reconocida en los mercados laborales, se debe a que los mercados laborales varían en su estructura, contexto social e historia, características que son particulares de cada mercado laboral, las cuales, es casi imposible que se repitan en dos o más regiones. Se encontró que la estructura del mercado laboral y el capital humano explica un porcentaje importante de la variabilidad del empleo a nivel municipal. Tercer ensayo titulado: Identificación de Distritos Industriales en México. Se utilizó un el modelo de radiación para la movilidad y la migración para estimar los flujos de viajes al trabajo entre municipios en México ante la ausencia de estos datos. Posteriormente se utilizó análisis de clúster jerárquico para determinar MLL. Utilizando como unidad geográfica de referencia los MLL obtenidos se utilizó la metodología propuesta por el Instituto Nacional de Estadística Italiano (ISTAT) para determinar cuáles de los MLL tenían características de Distrito Industrial y cuál era el agregado sectorial de manufactura preponderante en cada Distrito Industrial. Conclusiones. El posible desarrollo de políticas económicas según las circunstancias locales. Investigaciones referentes a la movilidad de los trabajadores son de mayor importancia porque permiten realizar estudios de corte socioeconómico y del comportamiento de los mercados de trabajo, con especial atención al desempleo. Permiten también llevar a cabo estudios comparativos del desempeño económico y pronósticos regionales, evaluar la competitividad y las disparidades entre regiones, e identificar territorios frágiles que requieren apoyos especiales. Las tres aplicaciones distintas se puede ver la potencialidad de las herramientas estadístico-econométricas para definir zonas relacionadas con la actividad laboral y la relevancia de la misma en distintos contextos (España, México). Los resultados obtenidos son útiles para decisores políticos o empresariales. ; Understand the territory from the functioning of the economy and, in particular, from the spatial behavior of employment. Through the formation of geographical areas for the reporting of labor statistics. By measuring and comparing mobility and coordination processes and expansion of local labor markets (MLL). Quantifying the impact of human capital and the labor structure in the variation of employment at the regional level. And geographically locating Industrial Districts. In the first essay entitled: Areas commutes in Spain for mutivariante statistical analysis. It was investigated as to form groups of certain geographical units from shared working relationships between municipalities. The sample was divided into nine strata. For each stratum the employment relationship between municipalities in two-dimensional contingency tables showing the number of commutes and the origins and destinations of people employed in each municipality was captured. Starting from here Hierarchical Cluster Analysis was applied to form Local Labor Markets (MLL). Later in order to measure and compare the mobility and processes of articulation and expansion of each MLL, the differential behavior between 2001 and 2011 four variables in each MLL obtained was analyzed: autonomy of supply autonomy demand, employment and number of workers. Given the evolution between 2001 and 2011 of these four variables each MLL was classified into four different behaviors. It was also found that the situation of surplus markets in jobs in 2001 Strata men, women and skilled manual workers, step into deficit in 2011. The only stratum showed a surplus behavior was that of agriculture. Second essay entitled: Evaluation of measured variables associated with education in commercial areas derived from economies of agglomeration and economic development location in Spain. In this essay was modeled with the approach of geographically weighted regression, the variation of employment in Spain between 2001 and 2011 in terms of human capital and labor structure that lived in each municipality of the sample in 2001. By using regression geographically weighted is got the answer to the geographic proximity of the surrounding municipalities influences, such information was summarized in maps. In order to understand the behavior of employment in terms of human capital and labor structure throughout the country. Indicating areas with greater potential business opportunities. The spatial heterogeneity recognized in labor markets, is because labor markets vary in their structure, social context and history, characteristics that are particular to each labor market, which is almost impossible to be repeated in two or more regions. It was found that the structure of the labor market and human capital accounts for a significant percentage of the variability of employment at the municipal level. Third essay titled: Identification of Industrial Districts in Mexico. one radiation model for mobility and migration to estimate flows commutes between municipalities in Mexico in the absence of these data was used. Hierarchical cluster analysis subsequently was used to determine MLL. Using as a geographical unit of reference the MLL obtained the methodology proposed by the Italian National Statistics Institute (ISTAT) was used to determine which of the MLL had characteristics of Industrial District and what was the leading manufacturing sector added in each Industrial District. Conclusions. The possible development of economic policies according to local circumstances. Investigations concerning the mobility of workers are most important because they allow cutting and socioeconomic studies of the behavior of labor markets, with special attention to unemployment. Allow also carry out comparative studies of economic performance and regional forecasts, assess competitiveness and disparities between regions, and identify vulnerable areas that require special support. Three different applications can see the potential of econometric statistical-defining work activity related areas and the relevance of it in different contexts (Spain, Mexico) tools. The results are useful for political and business decision-makers.

After stagnating for much of its postcolonial history, economic performance in Sub?Saharan Africa has markedly improved. Since 1995, average economic growth has been close to 5 percent per year. Has Africa finally turned the corner? This paper analyzes growth accelerations and decelerations-that is, country level deviations from long?run trend growth. Seen from this perspective, Africa's record of slow and volatile growth reflects a pattern of offsetting accelerations and declines, and much of the improvement in economic performance in Africa post 1995 turns out to be due to a substantial reduction in the frequency and severity of growth decelerations. The fall in economic declines since 1995 is largely due to better macroeconomic policies, but changes in such 'growth determinants' as investment, export diversification, and productivity have not accompanied the growth boom. Lack of change in these variables and the significant role played by natural resources in sparking growth accelerations suggest that Africa's growth recovery was fragile, even before the recent global economic crisis. The paper concludes by setting out four elements of a strategy that can help move Africa from fewer mistakes to sustained growth: managing natural resources better, pushing nontraditional exports, building the African private sector, and creating new skills.

After stagnating for much of its postcolonial history, economic performance in Sub?Saharan Africa has markedly improved. Since 1995, average economic growth has been close to 5 percent per year. Has Africa finally turned the corner? This paper analyzes growth accelerations and decelerations-that is, country level deviations from long?run trend growth. Seen from this perspective, Africa's record of slow and volatile growth reflects a pattern of offsetting accelerations and declines, and much of the improvement in economic performance in Africa post 1995 turns out to be due to a substantial reduction in the frequency and severity of growth decelerations. The fall in economic declines since 1995 is largely due to better macroeconomic policies, but changes in such 'growth determinants' as investment, export diversification, and productivity have not accompanied the growth boom. Lack of change in these variables and the significant role played by natural resources in sparking growth accelerations suggest that Africa's growth recovery was fragile, even before the recent global economic crisis. The paper concludes by setting out four elements of a strategy that can help move Africa from fewer mistakes to sustained growth: managing natural resources better, pushing nontraditional exports, building the African private sector, and creating new skills.

This paper provides an overview of the major current debates on infrastructure policy. It reviews the evidence on the macroeconomic significance of the sector in terms of growth and poverty alleviation. It also discusses the major institutional debates, including the relative comparative advantage of the public and the private sector in the various stages of infrastructure service delivery as well as the main options for changes in the role of government (i.e. regulation and decentralization).

Even nearly ten years of solid growth cannot guarantee long-term income convergence. The countries of the Latin America and Caribbean region (LAC), like other emerging economies, have benefited from a decade of remarkable growth and some income per capita convergence towards the United States and other industrialized countries. Yet, despite this recent progress, LAC still faces a significant per capita income gap with the developed world. The studies in this volume contribute to the ongoing debate on the reasons for this persistent income gap and the potential drivers of convergence, and propose some broad avenues for reform.

This report addresses two questions: what explains Ethiopia's growth acceleration?; and how can it be sustained? In brief, the authors find that Ethiopia's rapid economic growth, concentrated in agriculture and services, was driven by substantial public infrastructure investment and supported by a conducive external environment. To sustain high growth, three policy adjustments are presented: identifying sustainable ways of financing infrastructure, supporting private investment through credit markets, and tapping into the growth potential of structural reforms.

학위논문(석사)--서울대학교 대학원 :국제대학원 국제학과(국제통상전공),2019. 8. 김종섭. ; 본 연구의 목적은 글로벌 가치사슬 (global value chain, GVC), 혹은 여러 국가에 걸친 생산 단계의 분화 과정에 참여하여 특정 형태의 중간재 무역이 증가할 때, 숙련도가 다른 한국 노동자들의 임금에 차별적인 영향이 발생하는 가의 여부를 실증적 분석을 통해 검증하는 것이다. 최근 여러 개발도상국과 선진국에서 국가 내 불평등이 심화되는 현상이 관측되고 있으며, 이는 학계 및 정책입안자들 뿐만 아니라 일반인들의 주요 관심 대상이 되었다. 불평등이 정치적 안정성과 사회 통합에 영향을 미친다는 사실은 오랜 시간 인지되었다. 많은 국가 내에서 포퓰리즘과 시위가 늘어나는 등 정치적 갈등이 심화되고 있으며, 세계에서 경제 규모로 각각 1, 2위를 차지하는 미국과 중국 간의 무역 분쟁으로 현실화된 보호무역주의의 재등장은 오늘날 세계화와 불평등 사이의 상관 관계가 정치적으로 더욱 심각한 의의를 가진다는 것을 보여준다. 그러나 국제 생산 네트워크는 여러 국경을 넘나드는 글로벌 공급망 무역을 통해 촘촘하게 이어지는데, 이러한 글로벌 밸류 체인이 확산된 오늘날에는 관세, 쿼터, 그리고 기타 수입·수출 규제와 같은 비관세장벽의 비용이 더욱 높아졌다. 다시 말해 GVC 시대에서는 각 생산 단계를 거치며 수입 중간재가 국경을 여러 번 넘나들면서 관세의 비용이 누적·증폭되며, 전통적으로 내수형으로 여겨지는 농업과 서비스 같은 산업에 속한 생산과 고용 역시 해외 시장에 의존하는 경향이 커지는데, 이는 내수형 산업들조차 직접적으로 수출되는 제조업품 속의 부가가치로 체화되어 간접적으로 부가가치를 수출하기 때문이다. 따라서 과거 시대에 비해 무역 장벽은 고용과 임금에 더욱 부정적인 영향을 미칠 수 있을 뿐만 아니라, 양자간 무역의 직접적인 당사자 뿐만 아니라 간접적으로 국제 공급 사슬 무역에 참여하는 수많은 관련 국가와 산업들 모두에게 영향을 미칠 수 있다. 따라서 현대 사회에서 보호무역의 비용이 유래없이 높아진 만큼, 과연 그러한 정책의 밑바탕이 된 불평등 문제가 정말 무역에서 비롯된 것인지 정밀하게 연구하는 것은 아주 중요한 문제라고 할 수 있다. 1980년대와 90년대 초반까지 주류 경제학자들의 전반적인 의견은 무역이 불평등에 미친 영향이 미미했으며, 고숙련·저숙련 노동자들의 임금 격차가 벌어진 데에는 숙련 편향적 기술 진보와 같은 다른 요인들이 훨씬 중대한 효과를 미쳤다는 것이었다. 그럼에도 불구하고 세계화와 불평등의 관계에 대한 정책 입안자들과 일부 학계의 염려는 계속되어 왔으며, 특히 해외 아웃소싱 혹은 오프쇼어링과 임금 불평등의 관계에 대한 최근의 경험적 연구들은 여러 상반되는 결과들을 도출하였다. 한편, 글로벌 밸류 체인과 생산의 파편화가 확산된 상황에서는 무역의 잠재적인 숙련 편향적 효과를 새로운 GVC와 부가가치 무역 지수들로 연구하는 것이 중요하다. 이는 리카르도나 애덤 스미스 시대처럼 수출 속 부가가치가 거의 100% 국내에서 생산되는 것이 아니라 해외에서 수입한 중간재 혹은 다른 투입 요소가 차지하는 해외창출 부가가치 비중이 매우 커졌기 때문이다. GVC참여가 노동시장의 소득 재분배에 미치는 영향에 대한 최신의 경험적 연구들 역시 서로 상충되는 결과들을 내놓은 점에서, 더욱 정교한 방법론으로 다듬어진 실증 분석의 필요성이 제기된다. 특히 국제 공급 사슬 무역 속에 체화된 기술과 노동은 전통적 무역 이상으로 산업 고도화나 추가적인 노동 수요와 공급의 이동을 유발할 수 있기 때문에, 같은 GVC무역이라도 산업 혹은 국가에 따라 다른 영향을 미칠 수가 있다. 글로벌 가치 사슬에 가장 활발하게 참여하는 국가 중 하나인 한국의 사례가 중요한 또다른 이유는, 많은 경제학자들이 대학교와 같은 고등 교육에 투자를 해서 고숙련 노동의 비중을 높이는 것이 고숙련·저숙련 노동자 간의 임금 불평등을 해소할 수 있는 효과적 방안으로 제시하고 있고, OECD에서 가장 높은 비율의 고숙련 노동자를 보유하고 있는 한국의 경우 대학 교육 이수자의 지속적인 증가가 있었음에도 불구하고 임금 불평등이 해소되기는커녕 심화되었다는 점이다. 따라서 GVC와 임금 불평등의 구조를 연구하는 것은 GVC참여를 통해서 한국과 비슷한 방식으로 산업들의 기술 구조를 고도화하고자 하는 개발도상국들에게 좋은 참고가 될 수 있을 것이다. 한국의 예는 또한 선진국들에게도 중요한 의의를 가질 수 있다. 한국은 선진국 중에서 특이하게도 강건한 제조업 기반을 유지하고 있으며 반면에 서비스 산업이 상대적으로 낮은 비중을 차지하고 있다. 이런 산업 구조를 가지고 있음에도 GVC참여가 숙련 편향적인 효과를 보인다면, 최근 미국과 같은 선진국들이 보호무역을 통해 억지로 자국으로 (점점 낮은 부가가치를 차지하는) 생산·조립 단계 공정을 되돌리려는 "리쇼어링"을 유도하더라도 그들이 원하는 불평등의 개선 효과가 없을 수도 있다는 점을 함의한다. 산업 구조, 국가의 위치와 규모 등 수많은 요인에 따라 GVC참여가 노동 시장에 미치는 영향이 상이할 수 있는 바, 본 연구는 최근 축적된 국제 생산 분업에 대한 전반적인 선행 연구 분석과 함께 한국의 오프쇼어링, GVC관련 무역, 해외직접투자, 그리고 개발 및 산업 고도화 등의 다방면적인 질적 특성을 살펴봄으로써 이질적인 여러 종류의 GVC참여 방식이 국내 노동자들의 숙련도에 따라 임금에 어떤 상이한 영향을 미칠 수 있는지에 대한 가설을 설정한다. 본격적인 양적 회귀분석에 앞서 질적인 분석을 겸하는 이유는 GVC 무역 내에 체화된 업무와 숙련도를 알아야 노동 시장에 미치는 영향을 보다 정확히 파악할 수 있는 상황 속에서, 현재의 부가가치기준 무역 데이터조차 가치사슬 내의 정확한 산업 고도화 방향과 직무의 구성을 알기 어렵기 때문이다. 연구 가설들을 검정하기 위해 먼저 한국고용노동패널데이터 (KLIPS)에서 추출한 7,689명의 개인과 총 31,974개의 관측치로 이루어진 표본을 구성한 후 2018년 발표된 가장 최신 형태의 경제협력개발기구 (OECD) – 세계무역기구 (WTO) 부가가치 기준 무역 (TiVA) 지표들을 병합한다. 이 실증 분석 모형은 2005년부터 2015년까지 64개국간의 부가가치 무역을 추정하는 TiVA의 36개 산업 수준 지표들을 2009년부터 2017년까지의 개인 수준의 한국 노동자 데이터와 연결한다. 교육 수준으로 측정된 노동의 숙련 수준을 각 노동자가 속한 산업의 세 종류의 GVC 참여 지수 (총 참여율, 전방 참여도, 후방 참여도)와 함께 교차항에 넣어 상호 작용 효과 존재 여부를 살펴본다. 실증 분석을 위해 우선 변형된 Mincer 형태의 임금 모형에 종속변수인 각 개인 수준의 임금과 핵심 독립 변수인 노동 숙련도와 GVC참여율로 구성된 교차항과 함께 다양한 통제 변수와 고정 효과를 넣은 후, 패널 회귀분석을 실시한다. 이처럼 산업 수준 GVC 무역 지표를 개인 수준 임금 데이터와 통합시키는 방법론은 산업 수준 GVC 교역 지수를 산업 수준 임금 데이터와 연결 지은 기존 선행 연구에 비해서 동시적 인과관계로 인해 발생할 수 있는 내생성 편의 문제를 어느 정도 통제할 수 있다는 점에서 상당한 이점을 가진다. 실증 분석 결과 전반적으로 산업 수준에서의 GVC참여가 여러 숙련도로 나뉜 개인 노동자 수준의 임금에 유의미한 차등적인 효과를 보이는 것으로 나타났다. 우선 교차항을 고려하지 않았을 때 전방, 후방 및 총 GVC참여율 모두 다른 변수들을 통제했을 때에도 통계적으로 매우 유의미하게 임금을 높이는 것으로 보였다. 하지만 이와 동시에 GVC참여는 고숙련 노동자들에게 상대적으로 더욱 큰 긍정적 임금 효과를 주는 숙련 편향적 효과가 있는 것으로 나타났다. 여러 종류의 GVC참여 중에서도 전방 참여가 가장 큰 숙련 편향성을 나타내는 것을 드러냄으로써, 본 연구는 GVC참여의 종류를 구분하는 것이 매우 중요하다는 점을 확인하였다. 이는 한국의 노동 시장에 대한 선행 연구들이 거의 다루지 않은 부분일 뿐만 아니라, 최근에 세계 단위로 분석한 연구와 정 반대의 결과를 보여주기 때문에 기존 연구에 상당 부분 기여한다고 할 수 있다. 본 논문의 결과는 또한 동일한 형태의 GVC무역도 국가의 개별적 특성에 따라 체화된 숙련수준과 생산활동의 구성비에 따라 노동시장에 미치는 영향이 다를 수 있다는 점을 시사한다. 한편, 결과의 강건성 검증을 위해 다른 형태의 통제 변수와 모형, 그리고 대안적인 핵심 설명 변수로 시간 래그 변수와 총수출액 대비 부가가치 수출액의 비율(VAX Ratio)을 사용했을 때에도 전반적인 회귀 분석 결과는 유사하게 나오는 것으로 확인하였다. 본 연구는 한국의 경우 글로벌 공급 사슬 무역에 참여하는 것이 적어도 미시적인 수준에서 노동 시장에 숙련 편향적인 효과를 가져온다는 것을 밝히면서도, 동시에 모든 종류의 GVC참여가 노동자들의 전반적인 임금 수준에 긍정적인 영향을 미친다는 점을 보여줌으로써, 최근 불평등을 해소하는 정책으로 확산되는 보호무역주의는 최적의 해결책이 아니라는 경제학의 관점을 경험적 분석을 통해 확인하였다. 본 논문에서 무역과 노동 경제학이 가장 많은 부분을 차지하지만, GVC와 관련된 연구가 여러 학제간 교류가 활발한 간학문적인 분야라는 점과 최근의 무역 전쟁 및 불평등 문제가 정책적으로도 중대한 사안인만큼, 본 연구에 포함된 여러가지 이론 및 실증 분석의 결과들은 정치학, 국제관계학, 정치경제학, 사회학, 교육학, 행정학, 그리고 경영학과 같은 다양한 분야의 연구자들에게 유용한 결과를 제시한다. ; The main objective of this study is to elucidate how exposure to globalization in the form of participation in global value chains (GVCs), or the fragmentation of different stages of production across national and regional borders, has affected the wages of workers with different skill levels in the labor market of South Korea. The rise of income inequality within many developed and developing countries has once again captured the interest of academia, the public, and politicians. It has long been known that inequality affects political stability and social cohesion. Nowadays, political tensions run high in many nations, and as can be seen from various social phenomena such as the rise of populism, civil protests, and protectionism in the form of an ongoing trade war between the world's two largest economies, the U.S. and China, the potential relationship between globalization and inequality continues to have ever more serious political implications. However, the costs of trade barriers such as tariffs, quotas, and other non-tariff barriers such as import or export restrictions are now higher than ever, due to the importance of cross-border supply chain trade that links international production networks. In an era of GVCs, tariffs are escalated because inputs must cross borders multiple times, while production and employment in many seemingly domestic-oriented industries such as agriculture and services actually depend on foreign markets, because their value-added is indirectly embodied as inputs in manufactured exports. Thus, trade restrictions may lead to significantly greater negative impacts on wages and employment than in previous eras. Moreover, barriers aimed bilaterally at one country can affect numerous other countries that participate in production sharing. In light of the high costs of protectionism in the contemporary world, an examination of whether trade actually has adverse distributional effects is crucial. Until the 1980s and early 90s, the consensus of neoclassical economists was that trade only had a minor impact on inequality while skill-biased technical or technological change and other factors were far more important drivers of divergences in the income of high and low skilled workers. Nevertheless, public suspicion and concern over the relationship has been unabated, and more recent literature on the relationship between offshoring and income inequality has shown conflicting results. At the same time, the expansion of global value chains and fragmentation of production increases the importance of studying the potential effects of a skill bias in trade with new GVC and value added trade indicators, since nowadays foreign intermediate goods and services are significantly embodied in the final product exports of a country, unlike the age of David Ricardo or Adam Smith, when exports were only domestically produced. Empirical findings regarding the relation between GVC participation and its distributional impacts on labor have been mixed, furthering the case for continued empirical investigation. The case of Korea, one of the most heavily integrated developed countries in GVCs, is also important because many economists have suggested that more investment in the tertiary education of unskilled workers can alleviate income inequality, but Korea has been experiencing a rise in inequality in spite of having the largest proportion of high skilled workers among OECD countries when following ISCED classifications. As such, a careful examination of how GVCs affect wage inequality can provide useful insights for developing countries that want to consistently upgrade their industries akin to the path that Korea has followed. Likewise, Korea's case has important implications for developed nations: Korea is an outlier among developed nations because it has a remarkably robust manufacturing sector as compared to services, yet, the existence of a skill bias of global supply chain trade in spite of this may imply that current high income economies tempted to engage in protectionism to "re-shore" overseas production back into national borders (such as the U.S.) might not achieve the distributional results they intended. The lower value-added assembly stages of manufacturing coming back would not necessarily contribute to reducing inequality in the home country. A careful examination of the literature on the labor market impacts of international production sharing, as well as the qualitative characteristics of Korea's offshoring, GVC-related trade, foreign direct investment, and development - industrial upgrading trajectory are factored into the formulation of several hypotheses on how heterogeneous types of GVC participation might impact workers of different skills in Korea. This is to complement the limitations of value added trade data in showing the composition of business functions as well as direction of industrial upgrading, as finding the specific mix of tasks and skills embodied in GVC trade is crucial to understanding labor market impacts. To test these hypotheses, a panel data set consisting of 7,689 individuals and 31,974 individual-year observations is constructed by merging and matching data from the Korea Labor Income Panel Survey (KLIPS) with the updated 2018 version of the Organisation for Economic Co-operation and Development (OECD) – World Trade Organization (WTO) Trade in Value Added (TiVA) indicators, which are derived from the Inter-Country Input-Output (ICIO) database. This empirical model links the 36 industry-level indicators of TiVA, which covers 64 economies for 2005-2015, with the micro-individual level data of Korean workers from 2009-2017. The skill level of labor, measured in terms of educational attainment, is interacted with three different types of GVC participation indices (total, forward, and backward) of the respective industries in which the workers are employed each year. The wages of each individual worker, the dependent variable, are regressed on this product term of skills and GVC participation, using a variation of the Mincerian human capital wage equation along with various controls and fixed effects appropriate for this multi-dimensional panel data analysis. This approach of investigating the relationship between industry-level cross-border production sharing indicators on individual-level variables has a significant methodological advantage compared to many earlier studies using industry-level wage variables. Combining the two different levels can mitigate endogeneity concerns that may arise due to simultaneity bias. Overall, the findings of this study show that differences in GVC integration at the industry level indeed have heterogeneous effects on wages of individual workers classified in different skill groups. While all three types of GVC participation have positive effects on wages when controlling for other variables, the direction and magnitude of coefficients for each group of workers suggests the existence of a "skill-bias," in which increased GVC participation has a relatively favorable impact toward higher skilled employees as opposed to low or mid-skilled workers. This skill bias is strongest for forward participation, which underlines the importance of distinguishing between different types of GVC participation, a factor which was neglected in previous empirical studies combining sector-level GVC indicators with individual-level labor data. The fact that these results directly contrast with a recent cross-country study that found skill-biased effects for backward GVC trade rather than forward supply chain linkages, suggests that the country-specific business functions, skills, and tasks embodied within intermediate inputs trade affect the causal relationship between both types of GVC participation and labor market impacts, in line with this dissertation's analysis of Korea's specific position in GVC trade and development trajectory. Moreover, robustness checks show that the results are generally stable when estimated with complementary or alternative specifications of variables and models, including time lags and the Value Added Exports (VAX) ratio. At the same time, although there is a skill bias of global supply chain trade, this research shows that overall wages of workers are positively affected through all types of GVC trade, hence leading to the suggestion that the current protectionist sentiment spreading in the global economy is not the optimal answer to deal with inequality. Although the study mostly draws insights from and fills in the gap in contemporary international trade literature and labor economics, the multi-disciplinary relevance of the findings with respect to global value chains and within-country income inequality should be of interest to scholars and policymakers of many fields, including political science, international relations, political economy, sociology, educational studies, public policy, and business management among others. ; Abstract i Table of Contents v List of Tables and Figures vii I. Introduction 1 1. Background and Research Motivation 1 2. Overview of the Study 12 II. Literature, Theoretical Framework and Hypotheses 17 1. Literature Review 17 1.1. Global Value Chains (GVCs) 17 1.1.1 Concept, History and Terminology 22 1.1.2 GVCs and Value Added Trade Data 33 1.1.3 Importance of Trade in Value Added 42 1.1.4 The Role of Services 44 1.2. Labor Market Impacts of Trade and Offshoring 46 1.2.1 Benefits of Trade 51 1.2.2 Traditional Trade and Inequality 53 1.2.3 Trade in Tasks and Wage Effects 59 1.3 Traditional Proxy Measures of Offshoring Trade 64 1.3.1 Broad and Narrow Offshoring 64 1.4. Second Generation Offshoring Statistics 68 1.4.1 Vertical Specialization 70 1.4.2 VAX Ratio 73 1.5. The GVC Participation Index 76 1.5.1 Backward Participation (Foreign VA in Gross Exports) 78 1.5.2 Forward Participation (Domestic VA in Exports to third countries) 79 1.5.3 Total GVC Participation 79 1.5.4 Data Limitation: Absence of Business Functions and Tasks 81 1.6 Additional Labor Market Impacts of GVC Participation 83 1.6.1 The Smile Curve and Industrial Upgrading 85 1.6.2 Higher growth, development and productivity 93 1.6.3 Empirical Analyses on GVCs and Employment 96 1.6.4 Cross-country Analyses on GVCs and Wages 97 1.7 Korea and Global Value Chains 101 1.7.1 Korea's Prominent Role in GVC Trade 101 1.7.2 Factors underlying Korea's GVC participation 114 1.7.3 Korea's Export-Led Growth and Industrial Upgrading Path 116 1.7.4 Shifting to Higher Value Added Activities and Offshoring Assembly 122 1.7.5 Empirical Literature on the Labor Market Impacts of Globalization in Korea 128 2. Hypotheses Formulation 144 2.1. Model Predictions 144 2.2. Summary of Hypotheses 153 III. Data and Empirical Methodology 155 1. Data Sources and Sample 155 1.1. OECD-WTO Trade in Value Added (TiVA) Indicators 155 1.2. Korea Labor Income Panel Survey (KLIPS) 158 2. Econometric Analysis 161 2.1. Baseline Panel Regression Wage Equation Model 161 3. Variable Construction 163 3.1. Constructing Variables from KLIPS 163 3.2. Skills and Educational Attainment Variables 166 3.3. Constructing GVC Trade Variables from OECD TiVA 170 3.4. Alternative GVC Measures for Robustness Checks 172 3.5. Matching GVC Industries with KLIPS 174 IV. Results and Interpretation 177 1. Main Specification 177 1.1. Total GVC Participation 180 1.2. Forward GVC Participation 187 1.3. Backward GVC Participation 192 2. Further Robustness Checks 195 2.1. International ISCED Definition of Skills 196 2.2. Robustness to Endogeneity and Simultaneous Equation Bias 198 2.3. Individual Fixed Effects 199 2.4. Alternative Specifications: Time-Lagged GVC Trade Variables 202 2.5. Alternative Specifications: Value Added Export (VAX) Ratio 208 V. Conclusion 218 1. Contribution to Economics Literature 218 2. Contribution to Policy-Making and Other Academic Fields 222 3. Limitations and Suggestions for Future Research 223 List of References (Bibliography) 228 국문 초록 (Abstract in Korean) 260 ; Master

This study examines trends in school dropout at the upper secondary education level across Latin America over the past two decades, and attempts to identify factors influencing these rates. The methodology contributes to the existing literature by employing repeated cross sections of data to track the life cycle path of representative groups of individuals belonging to a birth cohort, by constructing and analyzing a synthetic data base of household survey data from 18 countries. A key finding is that while upper secondary enrollment rates increased in the region, the proportion of upper secondary age youth dropping out of school has remained persistently high, despite relatively favorable macroeconomic conditions. Furthermore, the study traces the moment in the life cycle at which the majority of dropout takes place to reveal differences between countries. Finally, to explain the trends in upper secondary dropout rates, the study examines the impact of three groups of factors: (i) shifts in the cohort size and socioeconomic composition of the population eligible for entering upper secondary education; (b) the macroeconomic environment and labor market opportunities; and (c) the returns to schooling. A series of regressions shows that an important factor that may be driving higher dropout levels has been the higher numbers of students from poor socioeconomic backgrounds reaching the upper secondary level. In addition, high returns to education have been a pull factor into the schooling system, while, especially in countries where the majority of youth dropout early (prior to upper secondary education), the data confirm an apparent substitution effect due to the opportunity cost of forgoing employment opportunities. Overall, the findings confirm the importance of policy makers' focus on upper secondary education across Latin America and suggest implications for focusing the policy agenda.

This paper presents new evidence on the patterns of cyclicality in the fiscal policy stance of developing and industrialized countries over a period of more than three decades covering 180 countries during 1980–2012. First, the paper considers issues of robustness in the choice of the proxy for fiscal cyclicality by using alternative filtering methods to check whether this influences the results and leads to any differences in a country's reported within-period average, and across-period changes in fiscal stance. Second, a country-specific approach is used to split the sample into sub-periods based on a test for structural break in the series of real gross domestic product per capita. Third, the paper investigates the extent to which countries behave pro-cyclically or counter-cyclically in different phases of the business cycle. In line with earlier findings in the literature, the analysis confirms that there is a causal link running from stronger institutions to less pro-cyclical fiscal policy, even after controlling for the endogeneity of institutions and other determinants of fiscal policy.

The purpose of this study is to demonstrate several techniques which can be used to evaluate pathways to sustainable growth in the Democratic Republic of Congo (DRC) via infrastructure improvement. Decades of conflict and neglect have left the DRC's transport infrastructure amongst the sparsest and most dilapidated in the world.

Trotz des erheblichen Ausmaßes der Bodenerosion in vielen Entwicklungsländern ist bislang weitgehend unklar, welches ihre wesentlichen anthropogenen Ursachen sind, und damit auch, wo Politiken und Maßnahmen für den Erhalt der Nahrungs- und Produktionsressource Boden ansetzen sollten. Jenseits unmittelbarer natürlicher und landnutzerischer Ursachen stehen heute sozioökonomische Faktoren im Mittelpunkt der Diskussion, von denen angenommen wird, daß sie die Anbau- und Bodenschutzentscheidungen der Landnutzer und darüber das Ausmaß an Bodenerosion beeinflussen, insbesondere: (i) verstärkte Armut, (ii) zunehmender Bevölkerungsdruck, (iii) verzerrte Agrarpreise, (iv) unangepaßter technischer Fortschritt sowie (v) unsichere Landbesitzverhältnisse. Der Bedeutung dieser Bestimmungsfaktoren wird vorwiegend im Rahmen produktionsökonomischer Ansätze und der Theorie der Induzierten Innovation nachgegangen. Allerdings wird die Wirkung einzelner Ursachen in der Literatur sehr unterschiedlich eingeschätzt. So wird beispielsweise in eher optimistischen Szenarien davon ausgegangen, daß Armuts- und Bevölkerungsdruck langfristig zur Entwicklung und Verbreitung bodenschonender Innovationen führen. In negativen Szenarien überwiegen hingegen Stimmen, die gerade in diesem Druck bei gleichzeitigem Preisdruck die wesentlichen Ursachen für die kurzsichtige Übernutzung des Bodens sehen. Empirische Studien zur Fundierung der kontrovers diskutierten Hypothesen liegen bislang nur für einen jeweils begrenzten lokalen Kontext vor und sind kaum verallgemeinerbar. Vor diesem Hintergrund bieten die Daten der ersten weltweiten Erhebung zum Stand der Bodenerosion (GLASOD, UNEP/ISRIC, 1991) nunmehr die Möglichkeit, sozioökonomische und landnutzerische Determinanten der Bodenerosion auf überregionaler Ebene empirisch zu untersuchen. Anhand der Aggregation und Analyse der im GLASOD enthaltenen Informationen wird zunächst deutlich, daß Afrika und Südostasien flächenmäßig mit jeweils rd. 4,5 Mio km2 am meisten zur Degradation durch Bodenerosion und Nährstoffverluste[1] in Entwicklungsländern beitragen, während der Anteil erodierter Fläche an der jeweiligen Gesamtfläche des Subkontinents[2] in Südwestasien (37%), Mittelamerika und Südostasien (jeweils rd. 25%) am höchsten ist. Extrem stark erodierte Länder finden sich v.a. in Mittelamerika und Afrika: In El Salvador, Haiti und Costa Rica sind zwischen 60% und 90% der jeweiligen Landesfläche betroffen. In Afrika sind vor allem die nord- und westafrikanischen Sahelländer Tunesien, Mauretanien, Libyen, Niger, Burkina Faso und Mali, im Osten die Hochlandstaaten Burundi und Rwanda sowie schließlich die Kapverdischen Inseln, besonders stark erodiert (40% bis 80% der Landesfläche). Wassererosion hat den größten Anteil an der Erosionsfläche, in Mittelamerika und Südostasien sind sogar mehr als 70% der erodierten Fläche von Wassererosion betroffen. Für die empirische Analyse der Zusammenhänge zwischen Bodenerosion und möglichen Bestimmungsfaktoren wird ein exploratives, ökonometrisches Vorgehen auf Grundlage nationaler Daten gewählt[3]. Die spezifische Aufeinanderfolge verschiedener Korrelations-, Faktoren- und Regressionsanalysen wird der großen Anzahl in Frage kommender Indikatorvariablen für mögliche Erosionsdeterminanten sowie den zu erwartenden Problemen der Multikollinearität und Modellspezifizierung in besonderem Maße gerecht. Letztere ergeben sich einerseits aus anzunehmenden Abhängikeiten unter verschiedenen Erosionsdeterminanten. Andererseits macht der latente Charakter[4], den die aus einem mikroökonomischen Kontext abgeleiteten Erosionsursachen auf aggregierter Ebene haben, es notwendig, für jede der angenommenen Determinanten verschiedene, u.U. korrelierte Indikatorvariablen zu definieren, was zusätzlich Kollinearität bedingt. Für Bodenerosion werden auf der Basis der national aggregierten GLASOD-Daten verschiedene Erosionsindizes definiert, die prinzipiell den von Wasser- und Winderosion sowie durch Nährstoffverluste betroffenen Anteil der nutzbaren Landesfläche wiedergeben. Die Datengrundlage für mögliche Erosionsdeterminanten wird ausgehend von Datensammlungen internationaler Organisationen für den Zeitraum 1961-1990 zusammengestellt. Für eine große Anzahl der in der Literatur diskutierten sozioökonomischen, landnutzerischen und auch natürlichen Rahmenbedingungen können repräsentative Indikatorvariablen definiert werden. Mangels geeigneter Indikatoren und Daten bleiben allerdings die Art und Sicherheit der Landbesitzverhältnisse unberücksichtigt. Insgesamt umfaßt die Datengrundlage rund 150 Variablen. Die Ergebnisse der Einfachkorrelationsanalysen zwischen den Erosionsindizes und möglichen Determinanten dienen einer ersten Einschätzung der Zusammenhänge. Sie zeigen, daß länderübergreifend insbesondere Variablen des Bevölkerungsdrucks sowie der durchschnittliche Waldanteil mit dem Ausmaß Bodenerosion in Zusammenhang stehen. Die Abholzungsraten in den 80er Jahren sind vor allem mit dem Ausmaß der Wassererosion korreliert. Bei Betrachtung der Länder mittleren Klimas[5] können Zusammenhänge mit Variablen nachgewiesen werden, die die Landnutzungsintensität und die Ausdehnung der tatsächlichen Nutzfläche in Relation zur potentiellen Nutzfläche wiedergeben. Weiterhin stehen in der mittleren Klimazone tendenziell sinkende Produzentenpreise für Agrarprodukte in Zusammenhang mit dem Ausmaß der Erosion. Erwartungsgemäß ist die Bedeutung natürlicher Faktoren für einzelne Erosionsformen und Klimazonen charakteristisch. Insgesamt scheinen Variablen, die das Ergebnis einer vermutlich längerfristigen Entwicklung wiedergeben, mehr Bedeutung für das Ausmaß der Erosion zu haben als solche, die Veränderungen im Referenzzeitraum 1961-1990 erfassen. Anhand verschiedener Faktorenanalysen für 62 Variablen und 73 Länder mit annähernd vollständigen Datensätzen können sodann strukturelle Zusammenhänge unter der Vielzahl möglicherweise relevanter Erosionsdeterminanten aufgedeckt und die Variablenanzahl auf Grundlage dieser Zusammenhänge auf eine geringere Anzahl weitgehend voneinander unabhängiger Größen reduziert werden. Es zeigt sich, daß die Struktur der Variablen durch etwa zehn gut interpretierbare Faktoren bei rd. 75% erklärter Gesamtvarianz klar wiedergegeben werden kann, und daß diese Faktoren auch bei Variation der Ausgangsvariablen sowie der Faktorextraktions- und Rotationsmethode stabil bleiben. Bemerkenswert ist, daß viele der Faktoren einen deutlichen Bezug zu den in der Literatur diskutierten Wirkungsketten unter möglichen Erosionsdeterminanten haben. So werden in dem für die Erklärung der Gesamtvarianz wichtigsten Faktor Variablen gebündelt, die die langfristige Intensivierung der Landnutzung im Zusammenhang mit strukturellem Bevölkerungsdruck und begrenzter Verfügbarkeit landwirtschaftlich nutzbarer Flächen erfassen. Weitere wichtige Faktoren beziehen sich auf strukturelle Armut in Verbindung mit erhöhtem ländlichen Bevölkerungswachstum; auf die mit Bevölkerungsdruck einhergehende langfristige wie auch rezente Expansion der landwirtschaftlichen Nutzfläche und Abholzung von Naturwald; auf Entwicklungswege, die eher auf die Produktion hochwertiger Produkte statt auf eine Flächenexpansion abzielen. Für die Preisentwicklung im Referenzzeitraum kann anhand einer Faktorenanalyse mit reduzierter Länderanzahl[6] gezeigt werden, daß ein Zusammenhang zwischen langfristig geringen oder negativen Preiszuwächsen im Agrarsektor und dem Faktor "Rezente Abholzungsraten" besteht. Um die relative Bedeutung dieser Faktoren für Bodenerosion zu quantifizieren, werden schrittweise Regressionsanalysen mit Bodenerosion als abhängiger Variablen und ausgewählten Repräsentantenvariablen für jeden Faktor als angenommenen unabhängigen Variablen durchgeführt[7]. Es lassen sich drei besonders relevante anthropogene Entwicklungen identifizieren, anhand derer das Erosionsausmaß bis zu rund 75% erklärt werden kann: (1) die langfristige, historische Ausdehnung der landwirtschaftlichen Nutzfläche auf Kosten des Waldbestandes in Zusammenhang mit einem Gesamtbevölkerungsdruck, der gegen Ende der 80er Jahre die agrar-ökologische Tragfähigkeit überschreitet; (2) die rezente Abholzung von Naturwald, die in Zusammenhang mit dem Wachstum der Gesamtbevölkerung zu sehen ist. Hier scheinen weniger der Druck der Agrarbevölkerung und die Ausdehnung der landwirtschaftlichen Nutzfläche - also die Produktionsseite - im Vordergrund zu stehen, als vielmehr der Druck der Nachfrageseite, in Kombination mit einem tendenziell sinkenden Agrarpreisniveau, das den Expansionsdruck auf das Land verstärkt hat. (3) Die langfristige, bevölkerungsdruckinduzierte Intensivierung der Agrarproduktion, vor allem durch Umwandlung von Dauergrünland in Ackerland, verkürzte Brachezeiten und erhöhte Viehbesatzdichten. Ein weiteres Ergebnis ist, daß in keinem Fall ein wesentlicher Einfluß von Armut auf das landesweite Ausmaß der Bodenerosion nachgewiesen werden kann - wie bereits die Ergebnisse der Einfachkorrelationsanalysen für immerhin 15 verschiedene Armutsindikatoren vermuten lassen. Es bestehen Unterschiede in den Erklärungsmustern für verschiedene Erosionsformen und Klimazonen. Die rezenten Abholzungsraten haben für Wassererosion, insbesondere in Ländern der extrem humiden Klimazone, herausragende Bedeutung. Zusätzlich zu den Faktoren (1) und (2) ist die Intensität der landwirtschaftlichen Produktion (3) vor allem für Wassererosion und in Ländern der mittleren Klimazone von Bedeutung. Hier ist auch die negative Wirkung einer sinkenden Agrarpreisentwicklung am stärksten. Gleichzeitig gilt hier: je eher der eingeschlagene Entwicklungsweg auf die Produktion hochwertiger Produkte im Gegesatz zur reinen Flächenexpansion abzielt, desto geringer ist das Erosionsausmaß. Für das Ausmaß der Winderosion und der Degradation durch Nährstoffverluste hingegen sind insbesondere die agroklimatischen Bedingungen ausschlaggebend. Die als erosionsrelevant identifizierten anthropogenen Rahmenbedingungen sind mit zentralen theoretischen Hypothesen konsistent. Fraglos gehören sie eher zu den Größen, deren kurzfristige Beeinflussung durch politische Maßnahmen schwierig ist. Dennoch können folgende Ansätze für eine Schwerpunktsetzung bei der Gestaltung von Politikmaßnahmen zur wirksamen Erosionsverminderung abgeleitet werden: Die Reduktion des Bevölkerungsdrucks durch eine an die natürlichen Bedingungen und relativen Faktorknappheiten angepaßte Erhöhung des Produktionspotentials, gerade auch in Regionen mit relativ niedrigem Potential. Eine stärkere Fokussierung auf Forstpolitiken bzw. auf eine Regulierung der kommerziellen Nutzung von Wäldern, vor allem in humiden Klimazonen. .Eine selektive, langfristig angelegte Verbesserung der incentive-Struktur für bodenschonende Produkte und Anbaumethoden über wirtschaftspolitische Eingriffe sowie durch verbesserte institutionelle und rechtliche Rahmenbedingungen. Von Politiken zur Armutsbekämpfung ist hingegen nicht zu erwarten, daß sie maßgebliche Impulse zur Verminderung der Bodenerosion geben können. Es muß jedoch immer präsent bleiben, daß arme Landnutzer sicherlich am stärksten und häufig existentiell von Erosionsschäden betroffen sind. Die Qualität zukünftiger Forschungsbemühungen auf globaler Ebene wird vor allem von der zukünftigen Datenverfügbarkeit und -qualität bestimmt: Für den Stand der Bodenerosion sind Informationen für verschiedene Zeitpunkte erforderlich; für anthropogene Erosionsdeterminanten eröffnen georeferenzierte Daten der Forschung gänzlich neue Perspektiven. Parallel zu überregionalen Analysen sind weitere lokale, sub-nationale Studien unbedingt notwendig, um umfassend zu ergründen, warum und welche Landnutzer die Ressource Boden in einem konkreten sozioökonomischen Kontext degradieren. Fußnoten: [1]Neben der Wasser- und Winderosion wird eine weitere Degradationsform, der Verlust von Nährstoffen und organischer Substanz, mitberücksichtigt und vereinfachend mit "Nährstoffverluste" bezeichnet.[2]Gemeint ist die nutzbare Landesfläche, Ödland ausgenommen. [3]Georeferenzierte Daten liegen derzeit für sozioökonomische Erosionsdeterminanten noch nicht vor.[4]D.h. Größen, von denen a priori nicht bekannt ist, wie sie beobachtet und gemessen werden können. [5]Dies sind Länder, in denen weder extrem aride noch extrem humide Bedingungen vorherrschen. [6]Für die entsprechende Variable liegen nur Daten für 56 Länder vor.[7]Umgekehrte Wirkungen der Erosion auf die als unabhängig angenommenen anthropogenen Variablen sind im Betrachtungszeitraum - bis auf die Armutswirkung starker Erosion - unwahrscheinlich. ; By the end of this century, soil erosion has reached an alarming extent in many developing countries. Still, uncertainty prevails regarding the human-induced causes of soil erosion. In consequence, many efforts to design efficient anti-erosion policies and instruments remain erratic. The actual discussion about human-induced causes of soil erosion focusses on socioeconomic factors that assumably influence the land users´ decisions on agricultural production and soil protection, and, hence, the degree of soil erosion. The most frequently discussed factors are: (i) poverty, (ii) population pressure, (iii) biased agricultural prices, (iv) the introduction of inadequate technical innovations and (iv) insecurity of land tenure. They are basically deduced from and discussed on base of production theory and the theory of induced innovation. Nevertheless, the different views on the importance to be assigned to the single factors are quite controverse. For example, in a rather optimistic scenario, it is argued that poverty and population pressure lead to the development of soil-conserving innovations in the long run. On the other side, poverty and population pressure, in combination with falling agricultural prices, are assumed to lead to a short-termist overuse of the soil. Empirical evidence that supports some of the controverse hypotheses on the causes of soil erosion is restricted to local studies based on local data on soil erosion, their results can hardly be generalized. In this context, the spatial data compiled within the global assessment of human-induced soil degradation (GLASOD; UNEP/ISRIC, 1991) for the first time permits a large-scale empirical analysis of socioeconomic and landuse factors relevant to erosion. By aggregating the information of the GLASOD data, countries and regions whith marked soil erosion can be identified. While Africa and Asia most contribute to the extent of soil erosion and the loss of nutrients[8] in absolute terms (4,5 mio sqkm each), it is in Southwest Asia (37%), Central America and Southeast Asia (25% each), where the proportion of of the land area - excluding wastelands - that is affected reaches the highest levels. Looked at on a national level, countries with an extreme extent of soil erosion are to be found in Central America and Africa: In El Salvador, Haiti and Costa Rica, 60 to 90 percent of the land area[9] are affected. In Africa, Sahelian Countries as Tunesia, Mauretania, Libya, Niger, Burkina Faso and Mali, as well as the eastafrican highlands of Burundi and Rwanda, and also Cape Verde show the highest proportions of eroded land area2 (40 to 80 %). Water erosion is the most widespread type of erosion, in Central America and Southeast Asia it even contributes with about 70% to the area affected by erosion and the loss of nutrients1. The methodological approach chosen for the empirical analysis of human-induced causes of soil erosion is an explorative, econometric one, based on national cross-country data[10]. A specific combination of correlation analyses, factor analysis, and regression analysis is designed, that can handle the great number of possible indicators for the assumed causes of erosion, and cope with related problems of multicollinearity and model specification. Those problems result from supposed interrelationships among different human-induced causes of soil erosion. At the same time, many of the causes of erosion have a latent character when considered on a national level[11], since they are deduced from a microeconomic context. This makes it necessary to define various indicator variables for each of them, which, again, implies additional multicollinearity. On the basis of the aggregated GLASOD data, a set of operational variables for soil erosion is defined. They basically indicate the proportion of a country´s degradable land area (i.e. land area minus wastelands) that is eroded through wind, water, or degraded by the loss of nutrients and organic matter by the end of the 80´s. In turn, the database for possible determinants of erosion is compiled departing from standard international data sets for the time span 1961-1990. Representative indicators can be defined for many of the causative factors discussed in literature, as well for socioeconomic ones, as for landuse, and also for natural factors. They are adapted in a way that they not only best fit and capture the hypothesized determinants, but also the ecological and timely dimension of the analysis. One important field that is not covered is land tenure. The resulting database comprises about 150 variables for possible causative factors, with a varying number of country-data available. The results of correlation analyses between the indicator variables for soil erosion and for possible causative factors facilitate a first assesion of relevant relationships. They show, that variables that quantify population pressure and the proportion of forested area are correlated with soil erosion for all countries. Deforestation rates in the 80´s are especially related to water erosion. Considering only countries without extreme climatic conditions[12] correlations are found between soil erosion and variables for the intensity of land use and the degree of expansion of the agricultural frontier. Producer price declines for relevant agricultural products are also found to be correlated with soil erosion in these countries. Corresponding to theoretical assumptions, the importance of different natural factors vary for different types of erosion and climatic zones. Altogether, variables that express structural conditions and can be regarded as the outcome of historical, long-term developments, seem to have stronger correlation with the extent of soil erosion than variables that quantify changes that took place within the time span under consideration, 1961 to 1990. The next methodological step consists in different factor analyses for 62 of the variables that express possible causative factors and for 73 countries with approximatively complete data sets. The principal objectives are to detect structural interrelationships among the multitude of variables and to reduce their number on the basis of these interrelations, in a way to obtain a set of variables that are largely independent of each other. It turns out that the structure of the 62 variables under consideration can clearly be reproduced by about 10 factors, with about 75% of their total variance being explained. These factors prove to be robust with respect to changes in the set of included variables, and in the methods of extraction and rotation. It is noteworthy, that many of the identified factors refer to cause-effect relationships that are discussed in literature. For instance, the factor that explains the greatest part of total variance, combines variables that quantify the long-run intensification of land use with others that stand for structural population pressure and a limited buffer for the expansion of the agricultural area. Other important factors relate to structural poverty, in combination with high rates of rural population growth; to the long-term and recent deforestation and to total population pressure; to development paths that aim at sopisticated animal procuction and permanent culture rather than at a mere expansion of the agricultural area. Other factors stand for the prevailing natural conditions. Based on a factor analysis for a reduced number of countries, it can be shown that declinig aggregate agricultural producer prices[13] are associated with the factor ´recent deforestation rates´. To quantify the relative importance of the identified factors, stepwise regression analyses are then carried out, with soil erosion as the dependent variable and selected representative variables for each of the factors as presumed independent variables[14]. Three human-induced factors, or developments, show to have particular relevance for the extent of soil erosion, that they can explain to up to 75%: (1) the long-run historical expansion of the agricultural frontier at the expense of the forested area, in combination with a population pressure well above the corresponding supporting capacities in the 80´s; (2) recent deforestation rates in conjunction with total population growth. This effect can rather be associated with a growth of demand for agricultural and forestral products and declining agricultural prices than with pressures directly resulting from agricultural population and expansion; (3) the long-run intensification of land use, mainly throug the conversion of permanent pastures to arable land, the shortening of fallow periods, and the increase of animal densities. This type of intensification is associated with and possibly induced by high structural population pressure in agricultural areas. Another important result is that poverty seems to have minor impact on the extent of soil erosion at the aggregate, national level. None of the included variables that represent the factor ´poverty´ shows a significant relative impact, neither in the models for the sum of erosion nor for specific types of erosion or climatic zones. This fact supports the low correlation coefficients for altogether 15 different poverty indicators that were calculated in the context of simple correlation analysis. Specific models for specific types of erosion and climatic zones show that there exist characteristic patterns of explanation for each type and zone. Recent deforestation rates and the associated features (factor (2))are particularily important in the explanation of water erosion, especially in countries with predominant humid climate. The impact of production-intensity in terms of factor (3) is specific for water erosion, and for countries without extreme climatic conditions, together with the factors (1) and (2). This is also where the negative effect of declinig agricultural prices appears to be strongest. At the same time, the development of sopisticated animal procuction and the growth of the area under permanent culture in contrast to a mere expansion of the agricultural area seem to be favourable to the soil in this context. In the explanation of wind erosion and loss of nutrients, natural factors are in the foreground. The identified, human-induced pressures related to long-term population growth, intesification, agricultural price decline and recent deforestation are consistent with important theoretical hypotheses. Those pressures are clearly not of the type that can be overcome over night through political intervention. Nevertheless, they lead to the following areas of intervention that should be given priority in the design of policy measures for the reduction of soil erosion: A reduction of population pressure through an increase in site-specific production potentials, based upon innovations that match the prevailing agro-ecological and economic conditions. Special attention should be given to low potential areas.A stronger focus on forest policy and the regulation of commercial forest use especially in the humid zone.A selective, long-term improvement of economic incentives for the production of soil-conserving crops with soil-conserving methods, by means of economic policy as well as through improved institutional conditions. Policies that aim at the reduction of poverty can not be expected to play a decisive role in the reduction of soil erosion. In spite of that, it is most necessary that policy makers keep in mind that the poor certainly are most affected by and vulnerable to erosion damages. At a global scale, the quality of future research on the topic will largely be determined by data availability and quality: concerning soil erosion, information at different points in time is necessary; for anthropogenic factors, spatial datasets will bring a new dimension into scientific research. Parallel with global analyses, further in depth local studies are necessary for a comprehensive and detailed insight into why and which land users degrade the resource they depend on in a specific socioeconomic context. footnotes: [8]The loss of nutrients and organic matter, independent of soil erosion, is also considered and is abbreviated with the term ´loss of nutrients´ in this text. [9]Again, it is the land area excluding wasteland that is being referred to. [10]Spatial data sets are not avaiable yet for socioeconomic factors related to soil erosion. [11]I.e. it is not known a priori, how these causes can be measured and quantified. [12]I.e. countries without predominant arid, hyper-arid or humid agroclimatioc conditions. [13]The availability of data for the variable in cause is limited to 56 coutries. [14]Reciprocal effects that soil erosion might have on anthropogenic factors are not very likely to occurr within the considered time span, except a possible increase of poverty due to erosion.