Research Review | 16 July 2021 | Forecasting

Forecasting the Long-Term Equity Premium for Asset Allocation
Athanasios Sakkas (U. of Nottingham) and Nikolaos Tessaromatis (EDHEC)
July 12, 2021
Long-term country equity premium forecasts based on a cross-sectional global factor model (CS-GFM), where factors represent compensation for risks proxied by valuation and financial variables, are superior, statistically and economically, from forecasts based on time-series prediction models commonly used in academia and practice. CS-GFM equity premium forecasts produce significant utility gains compared to long-term asset allocation strategies based on eighteen commonly used prediction models, consistently across the US and ten developed equity markets.

Interest Rate Skewness and Biased Beliefs
Mikhail Chernov (UCLA) and Michael Bauer (Universität Hamburg)
June 2021
Conditional yield skewness is an important summary statistic of the state of the economy. It exhibits pronounced variation over the business cycle and with the stance of monetary policy, and a tight relationship with the slope of the yield curve. Most importantly, variation in yield skewness has substantial forecasting power for future bond excess returns, high-frequency interest rate changes around FOMC announcements, and consensus survey forecast errors for the ten-year Treasury yield. The COVID pandemic did not disrupt these relations: historically high skewness correctly anticipated the run-up in long-term Treasury yields starting in late 2020. The connection between skewness, survey forecast errors, excess returns, and departures of yields from normality is consistent with a theoretical framework where one of the agents has biased beliefs.

Predicting VIX with Adaptive Machine Learning
Yunfei Bai (IEEE) and Charlie X. Cai (U. of Liverpool Management School)
June 14, 2021
Using 278 economic and financial variables we study the power of machine learning (ML) in predicting the daily CBOE implied volatility index (VIX). Designing and applying an automated three-step ML framework with a large number of algorithms we identify Adaptive Boosting as the best classification model chosen at the validation stage. It produces an average rate of 57% during the 11-year out-of-sample period. Potential significant economic gains are demonstrated in various applications with tradable instruments. Besides the modelling techniques, the weekly US jobless report is the most important contributor to the predictability along with some S&P 500 members’ technical indicators.

Forecasting GDP Growth Rates Using Google Trends in the United States and Brazil
Evripidis Bantis (ICMA Centre), et al.
June 7, 2021
In this paper we consider the value of Google Trends search data for nowcasting (and forecasting) GDP growth for a developed (U.S.) and emerging-market economy (Brazil). Our focus is on the marginal contribution of “Big Data” in the form of Google Trends data over and above that of traditional predictors, and we use a dynamic factor model to handle the large number of potential predictors and the “ragged-edge” problem. We find that factor models based on economic indicators and Google “categories” data provide significant gains compared to models that exclude this information. Using more disaggregated Google Trends data than its “categories” is not helpful. The benefits of using Google Trends data are greater for the U.S. than Brazil.

Forecasting US Inflation in Real Time
Chad Fulton and Kirstin Hubrich (Federal Reserve)
March 2021
We perform a real-time forecasting exercise for US inflation, investigating whether and how additional information–additional macroeconomic variables, expert judgment, or forecast combination–can improve forecast accuracy and robustness. In our analysis we consider the pre-pandemic period including the Global Financial Crisis and the following expansion–the longest on record–featuring unemployment that fell to a rate not seen for nearly sixty years. Distinguishing features of our study include the use of published Federal Reserve Board staff forecasts contained in Tealbooks and a focus on forecasting performance before, during, and after the Global Financial Crisis, with relevance also for the current crisis and beyond. We find that while simple models remain hard to beat, the additional information that we consider can improve forecasts, especially in the post-crisis period. Our results show that (1) forecast combination approaches improve forecast accuracy over simpler models and robustify against bad forecasts, a particularly relevant feature in the current environment; (2) aggregating forecasts of inflation components can improve performance compared to forecasting the aggregate directly; (3) judgmental forecasts, which likely incorporate larger and more timely datasets, provide improved forecasts at short horizons.

Modeling and Forecasting Macroeconomic Downside Risk
Davide Delle Monache (Bank of Italy), et al.
March 16, 2021
We document a substantial increase in downside risk to US economic growth over the last 30 years. By modelling secular trends and cyclical changes of the predictive density of GDP growth, we find an accelerating decline in the skewness of the conditional distributions, with significant, procyclical variations. Decreasing trend-skewness, which turned negative in the aftermath of the Great Recession, is associated with the long-run growth slowdown started in the early 2000s. Short-run skewness fluctuations imply negatively skewed predictive densities ahead of and during recessions, often anticipated by deteriorating financial conditions, while positively skewed distributions characterize expansions. The model delivers competitive out-of-sample (point, density and tail) forecasts, improving upon standard benchmarks, due to the strong signals of increasing downside risk provided by current financial conditions.

Forecasting Market Crashes via Machine Learning: Evidence from European Stock Markets
Hubert Dichtl (Dichtl Research & Consulting), et al.
May 10, 2021
This paper uses a comprehensive set of variables from the five largest Eurozone countries to compare the performance of simple univariate and machine learning-based multivariate models in predicting stock market crashes. The statistical predictive performance of a support vector machine-based stock market crash prediction model is significantly different from zero and among the best-performing univariate benchmarks, while still being truly out-of-sample. The ability to forecast subsequent stock market crashes out-of-sample translates into value-added to investors under realistic trading assumptions (net of transaction costs). Incorporating nonlinear and interactive effects is both imperative and foundation for the predictive performance of support vector machines. This adds an economic component to the advantageousness of machine learning-based multivariate crash prediction models over their univariate counterparts. It helps identify and explain the complex relationships in the underlying economic conditions (key economic drivers) that precede substantial stock market downturns.

Learn To Use R For Portfolio Analysis
Quantitative Investment Portfolio Analytics In R:
An Introduction To R For Modeling Portfolio Risk and Return

By James Picerno