بازبینی تجارت کردن نرخ تورم-بیکاری ایالات متحده : شواهد جدید برای سیاست گذاری

This paper addresses the various methodological issues surrounding vector autoregressions, simultaneous equations, and chain reactions, and provides new evidence on the long-run inflation–unemployment tradeoff in the US. It is argued that money growth is a superior indicator of the monetary environment than the federal funds rate and, thus, the focus is on the inflation/unemployment responses to money growth shocks. Structural vector autoregression (SVAR) and generalised method of moments (GMM) estimations confirm earlier findings in Karanassou et al., 2005 and Karanassou et al., 2008b obtained from chain reaction structural models: the slope of the US Phillips curve is far from vertical, even in the long-run, which implies that the nominal and real sides of the economy are symbiotic. In the light of the significant and robust long-run inflation–unemployment tradeoffs, policy makers should reconsider the classical dichotomy thesis.

The extent to which movements in output and unemployment depend on monetary fluctuations is a crucial issue in today's macroeconomics. Of particular interest, in this context, is to know how nominal magnitudes (such as the price level or money supply) interact with real economic activity and how this interaction evolves with the passage of time. The conventional wisdom supports the classical dichotomy and the absence of an inflation–unemployment tradeoff in the long run, i.e. the slope of the Phillips curve (PC) is vertical in the long run. Sachs (2009) points out that “The crash of 2008 exposed deep failures at the core of macroeconomic policymaking and macroeconomic thinking, and argues that the heated debates between neo-Keynesians and free-market economists have obscured a large and ultimately damaging consensus on economic thinking.” (Sachs, 2009, p. 1). We regard the classical dichotomy to have a prominent role in this consensus. The main objective of our work is to provide further evidence on the long-run relationship between inflation and unemployment for the US. This evidence is based on new estimates of the slope of the PC using two popular econometric techniques, structural vector autoregression (SVAR) and generalised method of moments (GMM), and supports previous results obtained through the estimation of dynamic multi-equation structural models. We refer, in particular, to the downward sloping PC documented for the US in Karanassou et al., 2005 and Karanassou et al., 2008b over the 1966–2000 and 1963–2005 periods. While we argue against the classical dichotomy doctrine, we unequivocally do not believe that it should be replaced by another doctrine. The existence of a downward-sloping PC at all time horizons simply demonstrates that money is an important driving force of unemployment. Nevertheless, in line with Ostrom (2007, p. 15176), we do not fall into panacea traps and do not perceive the inflation–unemployment tradeoff as a blueprint for reducing unemployment, since this would shift the focus of policy makers away from identifying the factors which jointly drive inflation and unemployment.1 As Phelps (2009a) argues we cannot understand well the big decline in aggregate employment without making allowance in our thinking for some important changes in the structure of the economy. The main contribution of our paper is to assess the robustness of the results in Karanassou et al., 2005 and Karanassou et al., 2008b using different empirical procedures. Furthermore, we provide a systematic compare and contrast discussion of the various methodological issues associated with the econometric approaches of structural models, SVAR and GMM, and highlight their respective pros and cons. This discussion is necessary in view of the common assertion that “There is as yet little certainty about how best to specify an empirically adequate model of aggregate fluctuations.” (Woodford, 2009, p. 275) In the context of dynamic multi-equation systems (like SVARs and “structural” models), the inflation–unemployment tradeoff can be measured by the ratio of inflation and unemployment responses to a monetary policy shock. In turn, in the context of single equations (like the hybrid new Keynesian PC model), this tradeoff is measured by the short- and long-run sensitivities of inflation to the unemployment rate. In both cases, the Phillips curve can be regarded as a function that, given a monetary policy shock, translates the impulse response function (IRF) of unemployment into the IRF of inflation and vice versa (Mankiw, 2001). Therefore, to evaluate the tradeoffs between nominal and real developments it is crucial to identify the relevant monetary policy shock. We proxy monetary policy by the growth rate of money supply because we believe money growth is a better indicator of the overall monetary conditions than the federal funds rate. This is a controversial issue, and the literature has argued both in favour of (Nelson, 2003, Nelson, 2008, Reynard, 2007 and Favara and Giordani, 2009) and against (Estrella and Mishkin, 1997, Woodford, 2003 and Woodford, 2008) the role played by money supply in the conduct of monetary policy. In Section 2 we argue that the monetary environment is better described by money growth than the widely used federal funds rate, since money growth reflects not simply the level of the yield curve but its slope and curvature as well. Furthermore, Nelson (2008, p. 1797), among other studies, defends the proposition that money growth does actually determine inflation in the long run. Consequently, in what follows we consider money growth shocks as the impulses that propagate the time-varying responses of the inflation and unemployment rates within a SVAR (or any dynamic multi-equation) model, and money growth as an instrument in the GMM estimation of the Phillips curve slope. Karanassou et al., 2005 and Karanassou et al., 2008b investigate the inflation–unemployment tradeoff using dynamic multi-equation models which feature spillover effects. They call their framework of analysis the chain reaction theory (CRT) and distinguish it from the traditional dynamic simultaneous equations (SE) framework which stems from the Cowles commission program. In contrast to the latter, CRT models focus on dynamics and flourish in a distributed-lag environment, placing emphasis on the role of IRFs. While dynamics and IRFs are also focal points in the (S)VAR framework, there are crucial differences between CRT and (S)VAR models that will be discussed in the course of the paper. As shown in Section 3, in a simultaneous equation model mere inspection of the individual equations only gives the short-run sensitivities2 of the endogenous variables with respect to the exogenous ones. CRT calls these sensitivities “local” and distinguishes them from the “global” ones, which are influenced by the inherent feedback mechanisms due to spillover effects (i.e. the simultaneity element)—“global” sensitivities can be adequately measured by the system's IRFs. The reason that CRT models refer to the inherent ‘simultaneity’ issue as ‘spillovers’ is to emphasise the plethora of feedback mechanisms contained in the system of equations, and flag their role in the measurement of the “global” sensitivities. The problem with the traditional SE estimates is that although they may seem reasonable if we take them at face value (“local” estimates), they might be misleading due to spillovers, in which case we need to look at the “global” estimates. In other words, although certain exogenous variables are significant in the respective equations of a labour market model, IRF analysis might reveal that they contribute minimally to the trajectory of the unemployment rate over a given sample period.3 We believe that a crucial factor which led to the disillusionment with the macroeconometric SE models, very popular in the past, was the lack of such a diagnosis. In sharp contrast to the SE models but in line with the CRT, such a diagnosis is a built-in feature of the (S)VAR methodology, since it relies on the IRFs of its closed systems. It is thus worth pointing out that, since the IRFs play a key role in dealing with the simultaneity element, the CRT can be regarded as an improvisation and synthesis of the SE and VAR approaches. Although both the CRT and (S)VAR procedures focus on the responses to impulses (shocks) and evaluate the “global” sensitivities of the variables under examination, they significantly deviate in how they depict an impulse: in the (S)VAR system shocks (impulses) arise from its error terms, while in the CRT system “shocks” refer to the actual changes in the exogenous variables. Therefore, an advantage of the CRT approach over the SE and (to a lesser extent) SVAR approaches is that the identification of policy effects is not a problem.4 We should emphasise that the CRT approach does not fall into the murky waters of the theoretical versus data-driven debate, since its models rely on the bidirectional feedback between a prior viewpoint and the observations-driven analysis. Put it differently, the CRT methodology investigates the interplay between theory and evidence, rather than compartmentalising them. In the empirical literature there is growing evidence that inflation and unemployment are interrelated in the long-run. For the US, Favara and Giordani (2009) estimate VARs on quarterly data and find that shocks to monetary aggregates contain substantial information on the future paths of output, prices and the interest rate. Ribba (2006) identifies a structural VECM and finds that permanent supply shocks explain the long-run comovement of inflation and unemployment in the US. These studies reinforce the findings of Campbell and Mankiw (1987) who estimate long-lasting real GDP responses to monetary disturbances using ARIMA models. Karanassou et al., 2005 and Karanassou et al., 2008b and Karanassou and Sala (2010a) apply the CRT methodology and find that the US inflation–unemployment tradeoff in the long run is between −3.5 and −3.7. This implies that a, say 10 percentage points (pp), increase in inflation (due to a permanent 10 pp increase in money growth) would reduce the unemployment rate by approximately 2.7–2.9 pp in the long-run. In the context of a SVAR model for Spain, Dolado, López-Salido, and Vega (2000) experiment with three alternative identifying assumptions to estimate the long-run tradeoff between inflation unemployment. In the monetarist scenario, where there is no long-run impact of supply shocks on the level of inflation, i.e. inflation is a demand (monetary) phenomenon in the long-run, they find that the slope of the long-run Phillips curve is −3.33. In the context of a CRT model, Karanassou et al. (2008a) find it to be flatter, around −2.7. Also for Spain, Bajo-Rubio, Díaz-Roldán, and Esteve (2007) estimate backward-looking Phillips curves, test endogenously for multiple structural breaks, and find a stable long-term tradeoff between inflation and the level of economic activity. For Italy, and also in a SVAR context, Ribba (2007) uncovers the influence of the monetary policy on the long-run movements in unemployment. In turn, Karanassou, Sala, and Snower (2003) use a CRT model for the EU and find that the slope of the long-run Phillips curve is −3.2. In the GMM literature of the new Phillips curve, it is common practice to restrict to unity the sum of the coefficients on the lead and lagged inflation terms. Since this is consistent with the conventional wisdom of a long-run vertical Phillips curve, not much attention is placed on testing such a restriction. In contrast, Karanassou et al. (2003) estimate a standard hybrid single-equation Phillips curve by GMM for the EU without imposing this a priori restriction. They find a long-run inflation–unemployment tradeoff between −3.1 and −3.5 (depending on the specific instrument list), which confirms their estimate of the long-run PC slope obtained through the CRT methodology. In Section 4 we contribute to the inflation–unemployment tradeoff literature by estimating the slope of the PC in the US using the SVAR and GMM methodologies. Our SVAR application is in line with the three-variable VAR model of Stock and Watson (2001) featuring inflation and unemployment, but with money growth instead of the federal funds rate. We estimate the PC slope by computing the impulse-response functions of inflation and unemployment to a money growth shock, and obtain a long-run tradeoff of −2.57. In turn, the GMM estimation of a standard hybrid specification of the new Phillips curve uses unemployment as the driving force variable and gives a long-run PC slope ranging from −3.30 to −4.32. Therefore, the SVAR and GMM estimates reinforce the robustness of a significant long-run inflation–unemployment tradeoff in the US. The rest of the paper is structured as follows. Section 2 discusses the advantages of money growth as an indicator of the overall monetary conditions, and overviews the robustness of the nonvertical PC within the CRT framework. Section 3 reflects on the salient features of the SE, SVAR and CRT methodologies, and uses an analytical illustration to uncover their differences and similarities. Section 4 presents the SVAR and GMM estimations of the long-run inflation–unemployment tradeoff. Section 5 concludes.

According to conventional wisdom an increase in the growth rate of money supply can have real effects only in the short run. In the long run, money growth will only increase inflation. This proposition of money superneutrality and the implied vertical long-run Phillips curve are being increasingly called into question. In this context, the value added of this paper is as follows. First, we argued in favour of evaluating the time-varying slope of the PC as the ratio of inflation and unemployment responses to a permanent money growth shock, and strongly supported the view that money growth is a superior indicator of the monetary environment than the federal funds rate. Second, we uncovered the various methodological issues surrounding the alternative dynamic multi-equation models of the PC: vector autoregressions, simultaneous equations, and chain reactions. Finally, we contributed to the growing empirical literature on the inflation–unemployment tradeoff by applying the widely used SVAR and GMM econometric techniques. In particular, we assessed the robustness of a downward-sloping long-run PC obtained by a chain reaction structural model for the US over the 1963–2005 period. We estimated the long-run tradeoff in the range of −2.57 and −4.32, which is in line with the findings of several studies for the US using different methodologies. Given the plethora of evidence against a vertical PC in the long-run, we conclude that policy makers should reappraise the classical dichotomy thesis. Although we believe that the recent financial developments strengthen the position that money growth is a better proxy of the monetary environment than the federal funds rate, a word of caution is required regarding the degree to which money growth can capture the overall monetary conditions. As shown in Fig. 3a, until the 3rd quarter of 2007 the rising money growth reflected the flattening of the yield curve and its eventual inversion (negative spreads). Nevertheless, since early-2008, money growth was unable to capture the exceptional circumstances of the financial crises (see Fig. 3a and b).In the light of the current economic crisis the Fed has tried to stabilise the financial system using both ‘qualitative’ and ‘quantitative easing’ (Bagus & Schiml, 2009). While qualitative easing refers to changes in the structure of the central bank's balance sheet, quantitative easing refers to its lengthening. In its first stage, until the end of summer 2008, the financial crisis was associated with a failing subprime market to which the Fed responded through qualitative easing. In its second stage following the Lehman Brothers’ bankruptcy, it became evident that the subprime crisis had mutated to a financial meltdown and the Fed resorted to quantitative easing as well. On one hand, in their efforts to boost the seriously troubled banking system and revive the credit markets, monetary policy makers have taken several innovative actions which cannot be fully captured by money growth (let alone short-term interest rates). On the other, as banks have been hoarding money to pump-up their balance sheets, quantitative easing has not been channelled into facilititating business activity. For example, Edlin and Jaffee (2009) note that from August 2008 to January 2009 excess reserves grew from $2 billion to $800 billion. We leave it to future research to explore how to quantify the recent unconventional monetary policies, the phenomenal rise of shadow banking in the 2000s, and their real impact on output and unemployment. Since the 2008 meltdown there has been an abundance of academic papers and articles in press on the ongoing debate about the need/efficacy of bailout plans (and the concomitant breach of free-market principles), what triggered the financial collapse, how to deal with the inflated budget deficits, the moral hazard implications of government interventions, the possibility of a return to a form of Glass–Steagall legislation (such as the “Volcker rule”), and the need/efficacy of reform and regulation of financial firm's compensation structures.21 Although these issues are beyond the scope of this study, our analysis should regarded as a stepping stone towards a better understanding of their implications for the real and nominal sides of the economy.