Skip to main content

Difficulties Forecasting Wage Growth

Wages have generated considerable discussion since the end of the recession. The income that households earn from working is an important support for consumer spending, which drives the bulk of activity in the U.S. economy. By this logic, strong labor income gains should boost consumer spending, thereby contributing to a strong economy, which begets strong hiring and wage gains, in a virtuous circle. Some previous research has found support for a wage Phillips curve: historically, as economic conditions have improved and the amount of slack in the labor market has decreased, wage growth has tended to pick up.

This business-cycle expansion has been notable because it has been characterized by a generally subdued rate of wage growth. Even though the unemployment rate fell from 9.8 percent to 5.5 percent between January 2010 and March 2015, growth in average hourly earnings for all employees on private nonfarm payrolls has been remarkably steady near a 2 percent annual rate. An alternative measure from the Bureau of Labor Statistics called the Employment Cost Index (ECI) captures broader compensation costs based on wages and salaries along with benefits. Growth in the ECI for compensation for private industry workers has been relatively similar.

Figure 1. Unemployment and Wage Growth

However, the far right side of the chart shows some positive signs. First, an unemployment rate of 5.5 percent is closing in on levels that many economists and policymakers think are consistent with relatively normal conditions. For example, in the Summary of Economic Projections following the March 2015 meeting of the Federal Open Market Committee (FOMC), the central tendency for the unemployment rate in the long run was 5.0 percent to 5.2 percent. Second, there are signs that compensation as measured by the ECI is accelerating. On a year-over-year basis, private industry compensation increased 2.8 percent through March 2015, its highest reading since September 2008. After a long stretch, the wage Phillips curve may finally be coming back to life.

A pickup in compensation growth is consistent with some reports coming from businesses. Several prominent national retail chains have recently announced plans to increase wages. The National Federation of Independent Business (NFIB) provides monthly survey evidence from small businesses showing the net percentage of respondents reporting plans to increase worker compensation in the next three months and the net percentage of respondents who increased worker compensation over the past three months. After falling off to extremely low levels during the recession, these readings have gradually recovered and are back within the range of readings from the previous two business cycles.

NFIB Compensation Measures

What can shrinking slack in the labor market or reports from businesses on their wage plans tell us about the trend for wages going forward? To address this question, I consider three models for forecasting wage growth. I take a broad view of “wages” by looking at employee compensation for private industry workers as measured by the ECI.

The first model is a medium-scale statistical model used in previous work, called a Bayesian vector autoregression (BVAR), which includes ECI growth, the unemployment rate, productivity, inflation, and several other typical macroeconomic data series. This model allows for the possibility that there is a wage Phillips curve in which a falling unemployment rate puts upward pressure on wage growth, but it also includes a variety of other factors that may affect wage growth, such as productivity and inflation. The second model uses information from businesses to predict future ECI growth. In particular, I map the NFIB survey responses on plans to raise worker compensation to ECI growth via a simple forecasting model. The third model is not much of a model at all: it simply assumes that future year-over-year ECI growth will be equal to its most recently observed value. This is a random walk model.

Using data available through the fourth quarter of 2014, I generate forecasts from these three models. After a decline in the middle of this year, the BVAR model puts ECI growth on an upward trajectory over the next several years, consistent with further improvements in labor markets, which the model predicts as well. By the end of 2017, ECI growth is a little above 3 percent in this forecast. The simple NFIB model almost perfectly predicted the ECI reading in the first quarter of 2015 of 2.8 percent. But this model would actually forecast that ECI growth should taper off somewhat, gradually falling to about 2½ percent by the end of the forecast period. By construction, the random walk forecast calls for ECI growth to be steady at a little under 2½ percent for the next three years. Of course, if I were to redo the forecasts using the most recent ECI reading of 2.8 percent, the random walk model would now call for that rate of ECI growth to persist going forward.

Given that the NFIB model made an excellent forecast for the first quarter of 2015, should we place the most weight on that model? Looking at the historical forecast accuracy of these three models is revealing. For each quarter starting in the first quarter of 1994 and ending in the fourth quarter of 2014, I generate the ECI growth forecast coming from each model for the next 12 quarters and then see how accurate those forecasts turned out to be. I assume the forecasts would have been made approximately in the middle of the middle month of the quarter, and wherever possible I use the data that would have been available to a forecaster in “real time” at that point. The BVAR model has historically generated reasonably accurate forecasts at short horizons and much less accurate forecasts at longer horizons, based on the typical forecast misses—technically, the root mean squared forecast errors—from this model. Relative to the BVAR model, typical forecast misses have been somewhat larger at short horizons for the simple NFIB model but smaller at longer horizons. But at each horizon, the random walk model has been the most accurate of the three models. This result suggests that movements in compensation growth—which depend on a complex combination of labor market slack, bargaining power, worker productivity, inflation, and myriad other factors—have been essentially unpredictable since the mid-1990s. These difficulties in forecasting labor compensation provide at least some evidence for why wages often appear to have little predictive power when forecasting inflation (see, for example, Stock and Watson 2008). In discussing the outlook for wages in her press conference following the March FOMC meeting, Federal Reserve Board Chair Yellen raised the possibility that wage growth may not pick up, a forecast in line with the predictions of a random walk model.

ECI Forecast Errors, 1994-2014

Of course, one distinct possibility is that these models for forecasting wage growth are inferior to other models. In this case, looking at the ECI forecast accuracy of other forecasters could be instructive. For the period 1994-2009, it is possible to see the publicly available forecasts for ECI growth that were made by one well-known forecasting body: the Federal Reserve Board of Governors staff, in the Greenbook. Greenbook forecasts are made immediately prior to each FOMC meeting. There are two regularly scheduled meetings of the FOMC in each quarter, and thus two Greenbooks; I use the second forecast from each quarter, potentially giving the Greenbook an information advantage over my previous forecasts, which were made using information available only through the first half of each quarter.

For the sake of comparability, I shorten the sample and look at the forecasting performance of the other models over the period 1994-2009 as well. Over short horizons—one to two quarters—the Greenbook’s forecasts for ECI growth were slightly more accurate than those from the other models. But as the forecast horizon lengthens, the typical forecast misses from the random walk model were again smaller than those coming from the Greenbook. In other words, extrapolating the recent past into the future was also a more accurate forecast for ECI growth on average than the Greenbook forecasts.

ECI Forecast Errors, 1994-2009

Upcoming EventsSEE ALL