Today, we are fortunate to present a guest contribution written by Paweł Skrzypczyński, economist at the National Bank of Poland. The views expressed herein are those of the author and should not be attributed to the National Bank of Poland.
Is it worth to use the U.S. Economic Policy Uncertainty Index (EPUI) as measured by Baker et al. (2016) together with the yield curve slope to track the business cycle turning points?
One can check how the EPUI performs when added to a standard term spread model specification to answer the above stated question. Monthly news-based EPUI time series goes back to 1985, however it is possible to construct a much longer time series of the EPUI with the use of historical data published at http://www.policyuncertainty.com. That is exactly what we perform here to obtain EPUI time series covering 1954-2018 period which allows us to estimate the augmented term spread model over the same sample as the standard term spread model. Namely we use monthly EPUI data covering Jan 1954 – Dec 2018 as shown in the figure below.
The analysis focuses on four specifications of probit models for the NBER dated recessions binary variable on the lefthand side of the estimation equation:
- spread_12 – model with 12-month lagged 10Y3M spread as the only regressor,
- spread_12_epui_6 – model with 12-month lagged 10Y3M spread and 6-month lagged EPUI as regressors,
- spread_6 – model with 6-month lagged 10Y3M spread as the only regressor,
- spread_6_epui_6 – model with 6-month lagged 10Y3M spread and 6-month lagged EPUI as regressors.
Tables below report the obtained estimates of model parameters.
Figures below present the obtained U.S. recession probabilities in-sample and their out-of-sample forecasts. Due to the structures of models used only the spread_12 model is able to forecast 12-months ahead, while remaining models forecast 6-months ahead. The predicted probabilities are reported in table 5.
The general picture that emerges is that adding the EPUI to a standard term spread model increases the volatility of probabilities, especially during this expansion, which is not surprising given that the EPUI itself became more volatile in recent years.
To check whether adding EPUI to the standard model makes sense we focus on the ROC curve analysis of the models. Table 6 reports false positive rates of the models for threshold probabilities from 10 to 90%. The indication here is not very appealing in favor of the augmented models as for 50% threshold models without EPUI are characterized by lower false positive rates.
Similarly, AUROC values, which can be interpreted as the goodness of fit measure of the models, indicate that the augmented models are very similar to the standard ones. Table 7 reports the obtained AUROCs. It is worth to outline that the null hypothesis of AUROC being equal to 0.5 is rejected at any reasonable significance level for each model, so we may conclude that all the models do better in predicting economic activity turning points than a toss of a coin does.
Overall once again the standard 10Y3M term spread probit model occurs to be a very strong benchmark when it comes to tracking U.S. business cycle turning points. The presented analysis shows that gains of including the EPUI in a standard model are quite scarce. In turn the answer to the question stated at the beginning is that using the EPUI to track business cycle turning points doesn’t seem to improve much the signals generated by the term spread itself. You can do it, but it doesn’t seem that it will help you much. Back to the drawing board then… Nevertheless, be aware that all models used here indicate that during 2019 the predicted odds of a recession in the U.S. are on the rise, which is associated both with the recently flattening yield curve and rising economic policy uncertainty.
Scott R. Baker, Nicholas Bloom, and Steven J. Davis. 2016. “Measuring Economic Policy Uncertainty.” Quarterly Journal of Economics, vol 131(4), pages 1593-1636.
This post written by Paweł Skrzypczyński.