Title: Back to the past: Long memory properties (again) matter for inflation forecasting
Authors: Todd Prono - Federal Reseve Board (United States) [presenting]
Abstract: Emerging from the high inflation period of the late-1970s and mid-1980s, the headline CPI YOY rate was characterized as a long memory process with hyperbolic decay. Over the next several decades, that long memory process degraded towards one of short memory with geometric decay. So stark was this degradation that over this period, the best out-of-sample forecasts, including multi-step-ahead forecasts, were produced by a model that ignores long memory properties altogether. By analyzing a series of multi-step, out-of-sample forecasts coming out of the COVD lock downs, it is demonstrated that long memory is back as a property affecting inflation dynamics. For instance, an ARFIMA model estimated over the high inflation period of the late-1970s and mid-1980s is shown to produce better medium-horizon CPI YOY rate forecasts than either an ARFIMA model estimated over a longer period that includes the Great Moderation or an ARIMA model estimated over the same period. An implication of this result is that the decay rate for headline inflation is time varying and slows down during periods of high inflation. Consequently, while the Great Moderation ushered in less `sticky' inflation dynamics, that lost `stickiness' has now returned.