Are Distributional Variables Useful for Forecasting With the Phillips Curve?

Sammanfattning: Does information on the distribution of wealth and income help us forecast aggregate macroeconomic variables? In this thesis, we study how adding such distributional variables to a standard forecasting model affects the forecast accuracy, in the context of inflation forecasting. Using the simulated inflation forecasting approach of Atkeson and Ohanian (2001), we perform a horse race between a textbook NAIRU Phillips curve to an extension augmented with variables from the wealth and income distributions. These are evaluated against a naive martingale model by comparing the root mean square errors (RMSE). Our initial results find that the top 1% wealth share slightly improves the RMSE:s compared to both the textbook and naive models. However, these results are not robust across different inflation measures. We also find that information on the income distribution does not improve the forecasts. In general, our results indicate no definitive improvements across all distributional variables. Lastly, our results are similar to those of Atkeson and Ohanian (2001), as we did not find the textbook NAIRU Phillips curve to have been helpful in predicting inflation over the past decade compared to the naive model.