Inflation and Unit Labor Cost Part 4

6. US INFLATION AND REAL UNIT LABOR COST 2000-2011 Thus far we have focused our attention on the 1960-1999 sample period that -- through the work of Gali-Gertler, Sbordone, Smets-Wouters and others -- played a key role in shaping the economics profession's views about the potential of the NKPE for explaining inflation dynamics. In the section, we extend the data through the 2000s to see how the NKPE instructs us to think about inflation during the last decade. This is interesting in its own right and also serves as an out-of-sample external validity check for the in-sample conclusions discussed earlier. Figure 5 plots real unit labor cost and modified real unit labor cost through 2011:I. While inflation has remained relatively stable over the past decade, both measures of real unit labor cost show significant declines (between 7 and 8 percent from 1999:IV to 2011:I). Of course, real unit labor cost (in logs, W – P + n − y) is labor's share of income in the sector: the stability of this "great ratio" is one of the standard stylized facts of macroeconomics. From a New Keynesian modeling perspective, the constancy is a production function attribute governing low frequency real reactions, but is overlaid by cyclical variations in ψ = W + n – P − y arising from the interplay of imperfect competition, sticky prices, and shocks. Over the benchmark sample period, 1960-1999, the average value of labor share in the non-farm business sector was 63.8%, while it was 65.1% in 1960.1 and 62.6% in 1999.4. In the last quarter of our data set, 2011.1, it had fallen to 57.8%. The source of this decline in labor's share (real unit labor cost) is both puzzling and important.21 However, in the context of the NKPE the puzzle about labor's share is not its source, but rather its predicted effect on inflation. Figure 6 plots inflation together with measures of fundamental inflation computed using the GG-VAR and Smets-Wouters-DSGE models extended with data through 2011:I. Evidently the large declines in real unit labor cost translate into large declines in fundamental inflation. While actual inflation is essentially unchanged over the post-1999 period, the measure of fundamental inflation constructed along Gali-Gertler lines fell by nearly 15 percent and that implied by the Smets-Wouters DSGE model fell by nearly 20 percent. That is, both measures of fundamental inflation predicted large deflation over the last decade. Our interpretation of this evidence is that real unit labor cost has exhibited important low-frequency variation that is unrelated to inflation. The last decade provides a dramatic example of this lack of comovement. However, our earlier analysis using the Smets-Wouters modified real unit labor cost suggests that the pattern of limited low frequency comovement also arose in the earlier sample period (1960-1999) and played out through the Smets and Wouters DSGE model in its estimation period. In this sense, their model was forced to deal with this low-frequency variability in the earlier sample period, and as we discussed above, the model did this by using the price-markup shock to (essentially) explain the trend real unit labor cost. Of course, given the structure of their model, any low-frequency variability in real unit labor cost must be intimately related to variability in inflation, so the same shocks that explained the trends in real unit labor cost (the price-markup shock) must also explain some of the variation in inflation (see Figure 8, panel A). Figure 14 decomposes our SW fundamental inflation measure over the extended sample period into a component associated with the model's price markup shocks and with all of the other shocks in the model. Given the analysis in the last section, it is not surprising that the price markup shocks explain the trend in fundamental inflation. However, it is interesting to note that the "detrended" version of fundamental inflation (that is, the component of fundamental inflation associated with shocks other than the price-markup shock) fits inflation much like our Gali-Gertler measure did over the 1960-1999 sample period. However, this component explains little of the variation in inflation since about 1985, following the disinflation of the early 1980s. 7. CONCLUSIONS An earlier generation of Keynesian macroeconometric models developed in the 1960s featured price and wage sectors that fit well during estimation periods, but that subsequently foundered on episodes of stagflation and the implication that there was an important long-run trade-off between inflation and unemployment. The Lucas critique led to an abandonment of these models by academics, a deepened skepticism about them within parts of the monetary policy community, and a search for macroeconomic models with stronger microeconomic foundations and better empirical performance. The microeconomic pricing foundations developed along New Keynesian lines incorporate expectations mechanisms which produce a shifting trade-off between inflation and real activity. They stressed the role of marginal cost in pricing and inflation dynamics, focusing on real unit labor costs as a proxy for marginal cost, and thereby avoided the thorny issue of measuring capacity output. Estimated over the mid 1960s through the early 2000s, these models provide an interpretation of the interval of rising inflation in the late 1960s and early 1970s, the subsequent stagflation episodes, the Volcker disinflation, and the Great Moderation beginning in the mid 1980s. The single equation studies of Gali and Gertler (1999) and Sbordone (2002) that popularized the New Keynesian Pricing Equation had three important properties that we display via replication and decompositions: a minor role of a transitory variation in marginal cost, a major role for expected of inflation, and an associated major role for a measure of expected future variations in marginal cost as captured by the construction of “fundamental inflation.” We show that the DSGE model of Smets and Wouters (2007), which contains a variant of the New Keynesian Pricing Equation, has the first two of these properties but is problematic along the third dimension: actual inflation and our SW fundamental inflation construction behave very differently. We trace this finding to the measure of real unit labor cost employed by Smets and Wouters (2007) which has a major downward trend so that their structural model is forced to include a major upward trend in a measure of price markup shocks. We considered how these structural markup shocks likely operate within the Smets-Wouters DSGE model, describing consequences for both real activity and inflation. Looking beyond the estimation period of the single equation studies and of the canonical DSGE model, we find that all measures of real unit labor costs display major downward trends that not accompanied by similar trends in inflation. (Equivalently, labor's share has fallen dramatically since the late 1990s, while there has been little variation in inflation). While it is possible to reconcile this absence of comovement by the introduction of large price markup shocks that are negatively correlated with real unit labor cost, we find it more plausible that real factors are influencing labor's share in ways that are largely unrelated to inflation. Tracing the sources of these changes and their implications for specification of pricing equations look to be first-order topics for macroeconomic research. Our work also illustrates a strategy which we believe will be an essential component of productive work with DSGE models, namely the detailed examination of a single structural equation or block of equations within using limited and full information from the complete model. The earlier generation of Keynesian macropolicy models were built by fitting together blocks of equations, whose performance was carefully scrutinized by the economists constructing these specifications. The builders of such earlier models were appropriately criticized in the late 1970s--by Robert Lucas, Thomas Sargent, Christopher Sims, John Taylor and others – for not paying sufficient attention to general equilibrium, including the consistency of behavioral specifications and the rationality of expectations. These critiques provided a major impetus for the development of modern DSGE models, including the now standard toolkit of linear rational expectations econometrics. Yet, DSGE modeling practice places nearly complete weight on exploration of system properties including measures of fit and forecasting performance, impulse response and moment analysis, and historical decompositions. With a focus on system properties alone, features that are important for central model components – such as the evolving joint behavior of inflation and real unit labor costs – are less likely to be recognized as central. However, detailed analysis of the structural inflation equation, as we have undertaken, makes imperfect specification more transparent. More generally, detailed exploration of equations and system blocks will lead to better DSGE model construction. Appendix A: Data All data are from FRED (Federal Reserve Bank of St. Louis). The table below documents the data series used in our analysis. Table A.1 Series FRED Label Desc Units Prices GDP deflator GDPCTPI Gross Domestic Product: Chain-type Price Index NFB Output deflator IPDNBS Household survey employment Employment NFB sector Population (16 and over) CE16OV Total Hours NFB HOANBS Nonfarm Business Sector: Implicit Price Deflator Employment Index 2005=100 Index 2005=100 Thousands Civilian Employment PRS85006013 Nonfarm Business Sector: Employment Civilian Noninstitutional Population (16 and older) : Note CNP16OV is a non-revised series that contains jumps, etc. We have eliminated outliers in growth rates and fitted cubic spline to monthly growth rates, then used this to get adjusted monthly levels, which were temporally aggregated to quarterly series. CNP16OV Nonfarm Business Sector: Hours of All Persons Avg. Weekly Hours NFB PRS85006023 Compensation per hour NFB sector − Nominal COMPNFB Real GDP GDPC96 Real Output NFB OUTNFB Investment − Nominal Consumption − Nominal GPDI Fed Funds Rate FEDFUNDS Nonfarm Business Sector: Average Weekly Hours Wages Nonfarm Business Sector: Compensation Per Hour Output, Consumption, and Investment Real Gross Domestic Product, 3 Decimal Nonfarm Business Sector: Output Gross Private Domestic Investment PCEC Personal Consumption Expenditures Interest Rates Effective Federal Funds Rate Index 2005=100 Thousands Index 2005=100 Index 2005=100 Index (2005 = 100) Billions of Chained 2005 Dollars Index 2005=100 Billions of Dollars Billions of Dollars Percentage points These data were used to construct two measures of real unit labor cost:    COMPNFB  HOANBS  The Gali-Gertler measure of log-real unit cost is GG = ln  .  IPDNBS  OUTNFB  The Smets-Wouters measure of modified real unit labor cost is given by   COMPNFB    PRS 85006023  CE16OV CNP160V a  SW  ln     t   ln     GDPCTPI     GDPC 96 1     ln  a    CNP160V     t   where CNP160Va is the adjusted value of population given in Table A.1, and  and  are parameters estimated by Smets-Wouters (  =0.00432 and  = 1.614980.) The data were also used to construct the seven variables used in the Smets-Wouters DSGE model. Using the notation in their equation (15), the variables are: GDP = GDPC96/ CNP160Va, CONS = (PCEC/ CNP160Va )/GDPCTPI , INV = (GPDI/ CNP160Va )/GDPCTPI, WAG = COMPNFB, HOURS = (PRS85006023/CE160V)/CNP160Va, P = GDPCTPI, and FEDFUNDS = FEDFUNDS. Appendix B: Additional Figures This appendix shows detrended versions of inflation and unit labor cost measures used by Gali-Gertler and Smets-Wouters. These detrended series were not used in our analysis, but the figures highlight the relationship between the variables absent their low-frequency/trend components. Figure B.1: Gali and Gertler Data Detrended Inflation and Real Unit Labor Cost Notes: This figure shows the detrended values of the series plotted in Figure 1. Detrending was carried out using a one-sided high-pass filter with a cutoff of 40 quarters. Figure B.2: Smets-Wouters Data Detrended Inflation and Modified Real Unit Labor Cost Notes: This figure shows the detrended values of the series plotted in Figure 3. Detrending was carried out using a one-sided high-pass filter with a cutoff of 40 quarters. Figure B.3: Comparison of Gali-Gertler and Smets-Wouters Measures of Real Unit Labor Cost Detrended Data Notes: This figure shows the detrended values of the series plotted in Figure 10. Detrending was carried out using a one-sided high-pass filter with a cutoff of 40 quarters.