George Akerlof - Behavioral Macroeconomics and Macroeconomic Behavior (Part IV : The Phillips Curve and the NAIRU) lyrics

[IV. The Phillips Curve and the NAIRU] Probably the single most important macroeconomic relationship is the Phillips curve. The “price-price” Phillips curve relates the rate of inflation to the level of unemployment, the expected rate of inflation, and variables affecting aggregate supply, such as the price of oil or food. The trade-offs between inflation and unemployment implicit in this relation define the “feasible set” for monetary policy and thus play a decisive role in its formulation. The Phillips curve was first estimated for Britain, then subsequently for the United States and many other countries. The basis of the Phillips curve is supply and demand. Phillips posited that when demand is high and unemployment low, workers can bargain for higher nominal wage increases than when demand is low and unemployment high. Firms' pricing policies translate wage inflation (adjusted for productivity) into price inflation. For policy makers, therefore, a durable trade-off exists between inflation and unemployment. In the late 1960's, Friedman (1968) and Phelps (1968) added an important new wrinkle. They argued that workers care about and bargain for real, not nominal, wage gains: workers routinely expect and receive compensation for expected inflation, then bargain from there, demanding higher expected real wage gains at lower rates of unemployment. Again, pricing policies translate wage inflation into price inflation. The consequence of this small shift in a**umption—that workers bargain for real, not nominal, wage increases—is enormous: instead of a durable unemployment-inflation trade-off, there is now just a unique “natural” unemployment rate consistent with stable inflation. With “real-wage” bargaining, the long-run Phillips curve—the unemployment/inflation combinations consistent with equality between actual and expected inflation—is vertical because there is one and only one unemployment rate: the “natural rate”—at which actual and expected inflation match.To see why the long-run Phillips curve must be vertical, imagine that a central bank attempts via monetary policy to hold unemployment below the natural rate. With labor markets anormally tight, workers demand nominal wage increases in excess of expected inflation (plus normal real wage cum productivity gains). Firms, in turn, pa** the a**ociated cost increases into prices, so that inflation exceeds what workers initially anticipated when they bargained. With unemployment below the natural rate, actual inflation therefore exceeds expected inflation. Ex post, workers have been fooled. So, over time, inflationary expectations, and inflation in turn, accelerates. With unemployment held below the natural rate, the consequence is ever accelerating inflation. Similarly, the Friedman-Phelps model predicts that a central bank attempting to hold unemployment above the natural rate indefinitely eventually causes accelerating deflation. Only the natural rate of unemployment yields steady inflation. Economists accepted the natural rate hypothesis remarkably quickly after it was first proposed by Friedman and Phelps in the late 1960's. Three things conspired in its favor. First, it seemed to explain remarkably well the inflation-unemployment experience of the 1960's and 1970's. At the low unemployment rates of the late 1960's, inflation rose, which apparently drove up inflationary expectations, shifting the short-run unemployment inflation trade-off outward. Thus the 1970's began with a much less favorable unemployment inflation trade-off than the 1960's. (an*lysts ignored the equally plausible explanation that as inflation increased, as it did in the late 1960's, wage bargains and price setting began to take inflationary expectations, which had previously been ignored, into account.) Second, empirical estimates of the Phillips curve yielded coefficients on past inflation whose sum was not statistically different from unity. The inference was drawn that the lagged inflation terms in such estimates correspond to expected inflation, which is an autoregressive weighted average of past inflation, and that the coefficient on expected inflation in determining current inflation is one. Finally, there is a bias for economists to accept rationally based null hypotheses, even though accepted only by tests with relatively low power. Economists should not have accepted the natural rate hypothesis so readily. There are both theoretical and empirical reasons to be highly suspicious. Theoretically, the natural rate hypothesis reminds me of a common diet book rule of thumb. According to that rule of thumb for every 3,200 calories extra that we eat, we gain a pound. For every 3,200 calories less, we lose a pound. This always makes me imagine twin brothers. One of these twin brothers eats just enough to keep his weight even. The other twin eats one more 100-calorie cookie per day. If the rule of thumb is right, after one year the cookie eater is 11 pounds heavier than his brother. After a decade he is 110 pounds heavier. Fifty years later, should he live so long, he would be 550 pounds heavier. Just as expected, the rule of thumb does break down when extrapolated over long time periods: more accurate renditions of the relationship between weight and calories show that the maintenance of higher weight requires extra caloric intake. Happily the twins' weights will not diverge forever. Similarly my guess is that for at least some band of unemployment rates, inflation would asymptote to a constant value rather than accelerate or decelerate indefinitely. Such a priori reasoning could be wrong, but the error from overextrapolation of the diet book rule of thumb warns us that the natural rate hypothesis is rather odd. At very low unemployment rates, the Friedman/Phelps prediction of accelerating inflation seems quite possibly reasonable and empirically relevant. But I am suspicious about the theory's applicability when unemployment is high. My suspicions regarding the natural rate hypothesis are supported by an empirical fact, which reveals that its applicability is not universal. Unemployment in the United States for the whole of the 1930's was indisputably in excess—surely greatly in excess—of any plausible natural rate. According to the natural rate hypothesis, price deflation should have accelerated for the whole decade. That did not happen. Prices fell for a time, but deflation stopped after 1932; there was no significant deflation for the next ten years, despite extremely high unemployment. This evidence suggests that, at least after some time, at high levels of unemployment and low inflation rates, the natural rate hypothesis breaks down. Such a failure would not be terribly serious for a theory derived from empirical observation, but it constitutes a serious flaw for a relationship derived from a priori principles, principles that are accepted because they are supposed to be always and everywhere true. The evidence of the 1930's is not unique. Modern economies display similar characteristics. For example, Pierre Fortin estimates that from 1992 to 2000, the Canadian economy experienced almost 12 points of unemployment in excess of a very conservative, 8-percent estimate of NAIRU. During that same period, inflation averaged a very low 11⁄2 percent per year. According to natural rate theory, core inflation should have declined by roughly 6 percentage points, since a typical estimate of the Phillips curve slope is 1⁄2. Instead, inflation declined over that period by only 0.1 percent. Econometric evidence further suggests that the natural rate theory rests on shifty sand rather than bedrock. Time-varying estimates of the natural rate show that it changes over time; but, even with allowance for such shifts, estimates of the natural rate possess high standard errors. Douglas Staiger et al. (1997) compute a 95-percent confidence interval for the U.S. natural rate which exceeds 5 percentage points; this is more than three times the standard deviation of the U.S. monthly unemployment rate over the last 50 years. In recent papers, William Dickens, George Perry, and I have explored two behavioral hypotheses that, contrary to the natural rate model, produce a stable trade-off between unemployment and inflation at sufficiently high unemployment and low inflation rates. The first hypothesis is “pure Keynes”: workers resist, and firms rarely impose, cuts in nominal pay. The second hypothesis concerns the role of inflationary expectations in wage bargains: we argue that, at very low inflation, a significant number of workers do not consider inflation sufficiently salient to be factored into their decisions. However, as inflation increases, the losses from ignoring it also rise, and therefore an increasing number of firms and workers take it into account in bargaining. Keynes' a**umption that workers resist nominal wage cuts was consistent with his intuitive understanding of psychology. The a**umption also coincides with psychological theory and evidence. Prospect theory posits that individuals evaluate changes in their circumstances according to the gains or losses they entail relative to some reference point. The evidence suggests that individuals place much greater weight on avoiding losses than on incurring gains. Daniel Kahneman and Amos Tversky (1979) have demonstrated that many experimental results which are inconsistent with expected utility maximization can be rationalized by prospect theory. Downward wage rigidity is a natural implication of prospect theory if the current money wage is taken as a reference point by workers in measuring gains and losses. In support of this view, Eldar Shafir et al. (1997) found in a questionnaire study that individuals' mental frames are defined not just in the real terms hypothesized by cla**ical economists but also exhibit some money illusion. Numerous empirical studies document that money wages are, in fact, downward sticky. Using panel data, David Card and Dean Hyslop (1997) and Shulamit Kahn (1997) found that distributions of nominal wage changes are asymmetric around zero. Fortin found a remarkable pileup of wage changes at zero in Canadian data. From 1992 to 1994, when Canadian inflation was 1.2 percent and the unemployment rate averaged 11.0 percent, only 5.7 percent of nonCOLA union agreements had first-year wage cuts, whereas 47 percent had wage freezes.47 In detailed interviews in Connecticut, Bewley found that managers are willing to cut nominal wages only as a last resort. To investigate whether firms cut total compensation through benefit cuts as opposed to money wage cuts, David E. Lebow et al. examined the individual industries covered by the Employment Cost Index: they found that benefit cuts are only a minor substitute for nominal wage cuts. Using Swiss data, Fehr and Lorenz Goette found that even a seven-year period of low inflation and low productivity growth did not increase the frequency of money wage cuts. At low inflation there is a long-run trade-off between output and inflation if there is aversion to nominal pay cuts. Unlike the FriedmanPhelps model, in which such a trade-off is transitory, long-term increases in inflation (if it is close to zero) result in significantly less employment and more output. The logic goes as follows. In both good times and bad, some firms and industries do better than others. Wages need to adjust to accommodate these differences in economic fortunes. In times of moderate inflation and productivity growth, relative wages can easily adjust. Unlucky firms can raise the wages they pay by less than the average, while the lucky firms can give above-average increases. However, if productivity growth is low (as it was from the early 1970's through the mid-1990's in the United States) and there is no inflation, firms that need to cut their real wages can do so only by cutting the money wages of their employees. Under realistic a**umptions about the variability and serial correlation of demand shocks across firms, the needed frequency of nominal cuts rises rapidly as inflation declines. An aversion on the part of firms to impose nominal wage cuts results in higher permanent rates of unemployment. Because the real wages at which labor is supplied are higher at every level of employment when inflation is low, the unemployment rate consistent with stable inflation rises as inflation falls to low levels. Spillovers produce an aggregate employment impact which exceeds the employment changes in those firms that are constrained by their inability to cut wages. Thus, a benefit of a little inflation is that it “greases the wheels of the labor market.” Simulations of a model with intersectoral shocks and aversion on the part of firms to nominal wage cuts suggests that, with realistically chosen parameters, the trade-off between inflation and unemployment is severe at very low rates of inflation, when productivity growth is low. For example, a permanent reduction in inflation from 2 percent per year to zero results in a permanent increase in unemployment of approximately 2 percentage points. Estimation of a Phillips curve for the United States after World War II, corresponding to the simulation model just described, gives similar results. When the Phillips curve thus estimated is used to simulate the inflation experience of the 1930's, the fit is shockingly close to actual U.S. inflation experience during the depression. A comparable simulation of the standard natural rate model, in contrast, counterfactually, shows accelerating deflation throughout the 1930's. An alternative behavioral theory also generates a permanent trade-off between inflation and unemployment at low inflation. This theory is based on the idea that because inflation is not salient when it is low, anticipated future changes in the price level are ignored in wage bargaining. With monopolistic competition and efficiency wages such ignorance of inflation when it is low is near-rational.55 The psychology of just noticeable differences and cognitive psychology both suggest that people tend to ignore variables that are unimportant to their decisions. Econometric estimates of the Phillips curve which allow for the possibility that past inflation has a different impact on current inflation when inflation is high than when it is low are consistent with this hypothesis: at high inflation, the sum of coefficients on past inflation is close to one. At low inflation, this sum of coefficients is much closer to zero. Similarly, regressions using survey measures of expected inflation as an independent variable yield much higher coefficients on the expected inflation term at high inflation than at low inflation. Not surprisingly then, when periods of low and high inflation are combined to estimate a nonlinear model of the influence of inflationary expectations we find that their impact depends on the recent history of inflation. The demonstration by behavioral macroeconomics that very low inflation has the cost of permanently high unemployment and low output, has important implications for monetary policy. Most of us think of central bankers as cautious, conservative, and safe. But I consider many to be dangerous drivers: to avoid the oncoming traffic of inflation, they drive on the far edge of the road, keeping inflation too low and unemployment too high. During the 1990's, Canada had very low inflation and an unprecedented unemployment gap—close to 4 percentage points—with the United States. Europe has also had high unemployment and very low inflation. Japan has gone much further, allowing deflation. Central bankers who accept the textbook version of the natural rate hypothesis should follow the advice of Oliver Cromwell to the General Assembly of the Church of Scotland: “I beseech you in the bowels of Christ, think it possible you may be mistaken.” It is no coincidence that the leading survey of cognitive psychology uses this citation to demonstrate a common perceptual error: overconfidence.