The Ins and Outs of Forecasting Unemployment: Using Labor Force Flows to Forecast the Labor Market

Abstract

This paper presents a forecasting model of unemployment based on labor force flows data that, in real time, dramatically outperforms the Survey of Professional Forecasters, historical forecasts from the Federal Reserve Board's Greenbook, and basic time-series models. Our model's forecast has a root-mean-squared error about 30 percent below that of the next-best forecast in the near term and performs especially well surrounding large recessions and cyclical turning points. Further, because our model uses information on labor force flows that is likely not incorporated by other forecasts, a combined forecast including our model's forecast and the SPF forecast yields an improvement over the latter alone of about 35 percent for current-quarter forecasts, and 15 percent for next-quarter forecasts, as well as improvements at longer horizons.1. A long literature discusses the apparent asymmetry of unemployment: for example, Mitchell (1927), Neftçi (1984, corrected by Sichel 1989), Rothman (1991, and more recently, Hamilton (2005), Sichel (2007), and Rothman (2008).2. Moreover, unlike standard time-series models used to capture asymmetries (such as threshold models or Markov switching models), our model does not rely on a threshold (which may change) to introduce asymmetry. A comprehensive study by Stock and Watson (1999) concluded that linear models generally dominated nonlinear models for out-of-sample forecasts of most macroeconomic time series. The unemployment rate is an exception; see, for example, Milas and Rothman (2008).

Heterogeneity and Unemployment Dynamics

“…8 For the estimation of univariate models, ARIMA(2,0,1) and ARIMA(2,1,0), I used rolling regression with a fixed window only. Following Barnichon and Nekarda (2012), I also compute the forecasts using their nonlinear steady-state model with constant hazards (s t and f t ), which does not require rolling regression of any type.…”

“…random variables with mean 0 and variance and ARIMA(2,1,0) as univariate forecasting models, the two specifications commonly used in the literature (Montgomery et al 1998;Barnichon and Nekarda, 2012).…”

“…The last class of models that I consider is a nonlinear steady-state model proposed by Barnichon and Nekarda (2012). They consider information from the labor force flows through the concept of the conditional steady-state unemployment rate.…”

“…horizons, from the current to 4-quarter ahead (Table 3.3). In addition, their RMSE's 13 Barnichon and Nekarda (2012) also used this approach. 14 The followings are variables going into the vector x t in the forecasting models: (1) VAR1: ln s t , ln f t , (2) VAR2: ln uic t , d ln hwi t , (3) VAR3: ln uic t , (4) VAR4: exit probabilities of those who have been unemployed for 1 month and 2-3 months.…”

Kalman Filter or VAR Models to Predict Unemployment Rate in Romania?

“…A detailed study regarding unemployment forecasts and predictions performance carried out by Barnichon and Nekarda (2012) Heilemann and Stekler (2013) offered several reasons for the lack of accuracy of G7 predictions in the last 50 years. They identified one continuous critique brought to macro-econometric models and forecasting techniques, but also concluded that the accuracy expectations are not realistic.…”