PREDICTING SYSTEMATIC RISK WITH MACROECONOMIC AND FINANCIAL VARIABLES

• Author: Tolga Cenesizoglu,Denada Ibrushi
• DOI: http://doi.org/10.1111/jfir.12221
• Date: 01 August 2020
• Published date: 01 August 2020

The Journal of Financial Research Vol. XLIII, No. 3 Pages 649–673 Fall 2020

DOI: 10.1111/jfir.12221

PREDICTING SYSTEMATIC RISK WITH MACROECONOMIC AND

FINANCIAL VARIABLES

Tolga Cenesizoglu

HEC Montreal

Denada Ibrushi

St. Mary’s University

Abstract

We analyze the predictive power of several macroeconomic and financial indicators

in forecasting quarterly realized betas of 30 industry and 25 size and book‐to‐market

portfolios. We model realized betas as autoregressive processes of order 1 and

include lagged values of macroeconomic and financial indicators as exogenous

predictor variables. In out‐of‐sample forecasting exercises, forecasts using bond

market variables as exogenous predictors statistically outperform forecasts from a

benchmark model without any exogenous predictors. These forecasts based on bond

market variables also economically outperform benchmark forecasts by providing

better performance in hedging the market risk of portfolios.

JEL Classification: C10; G17

I. Introduction

Accurate beta forecasts are of central importance for many applications in finance

including but not limited to asset pricing, cost of equity estimation, and portfolio

management. For example, hedge funds frequently employ market‐neutral strategies,

which are supposed to provide consistent returns independent of market fluctuations.

However, it has been shown that returns of these hedge funds are exposed to market

risk (Asness, Krail, and Liew 2001; Patton 2009; Bali, Brown, and Caglayan 2014).

For instance, Bali, Brown, and Caglayan (2014) document a positive and significant

relation between hedge fund returns and market returns even after taking into account a

variety of control variables. Hence, the success of these so‐called market‐neutral hedge

funds depends crucially on obtaining good beta forecasts.

In this article, we propose using several macroeconomic and financial

indicators as exogenous predictor variables to obtain better beta forecasts. We do this

by analyzing whether certain macroeconomic and financial indicators provide any

additional power in predicting the quarterly realized betas of 30 industry and 25 size

We thank Mathieu Fournier, Gunnar Grass, Marie Lambert, and Harald Lohre for their helpful comments.

We also want to thank the participants at the 2018 Southern Finance Association annual meeting for the very

constructive feedback that was subsequently implemented in our article.

649

© 2020 The Southern Finance Association and the Southwestern Finance Association

and book‐to‐market portfolios. We consider 30 variables and group them into six

categories: aggregate corporate variables, equity market variables, bond market

variables, aggregate valuation ratios, macroeconomic variables, and portfolio‐specific

variables.

We obtain time series of quarterly realized betas by regressing daily (excess)

portfolio returns on daily (excess) market returns in each quarter between 1929 and

2016. We first consider several benchmark models to capture the time‐series dynamics

of these realized betas. We find that an autoregressive (AR) model of order

1 outperforms historical mean, random walk, and Fama–MacBeth (1973) approaches in

forecasting realized betas out of sample. Hence, we consider the AR model as our main

benchmark and include lagged values of each macroeconomic or financial variable

separately as exogenous predictor variables in this model. We refer to these

autoregressive models with exogenous variables as ARX models. Comparing the

performance of these ARX models and the benchmark AR model without any

exogenous variables allows us to analyze the predictive power of macroeconomic and

financial variables beyond the own dynamics of realized betas. We estimate the ARX

model using a rolling window of 60 quarters and obtain one‐step‐ahead out‐of‐sample

forecasts for each portfolio and predictor variable. Rather than considering the

forecasts based on each predictor variable separately, we use a forecast combination

approach by considering the mean and median forecasts in each category. This

approach not only increases the accuracy of our forecasts but also allows us to

understand which variables as a group are informative about future betas. We evaluate

the out‐of‐sample performance of these forecasts in three samples: 1929–1964 (early

sample), 1965–2016 (modern sample), and 1929–2016 (whole sample).

Although our main interest is the performance of ARX models relative to the

AR model, it is also important to understand absolute performance of these approaches

and analyze whether they provide unbiased and efficient forecasts of realized betas. We

test this based on the approach proposed by Mincer and Zarnowitz (1969). Overall, our

results suggest that the benchmark AR model as well as its variant ARX models

are reasonably good models for realized betas of different portfolios in the modern

sample but not necessarily in the early sample. This in turn implies that our results on

the relative performance of different models in the early sample should be interpreted

in light of their relatively poor absolute performance in this sample.

We turn our attention to their relative statistical performance. We first consider

their relative statistical performance based on their mean squared errors (MSEs). The

ARX model outperforms the benchmark AR model for most of the variable categories

across portfolio groups and samples. Furthermore, the mean and median forecasts of all

individual ARX models provide significantly lower MSEs than the AR model

consistently across portfolio groups and samples. These results in turn suggest that

macroeconomic and financial variables provide some useful information in predicting

systematic risk beyond the lagged values of realized betas, at least in the statistical

sense. We also find that only two categories—bond market and portfolio‐specific

variables—provide consistently better forecasts than the benchmark AR model across

different portfolio groups and samples. Furthermore, portfolio‐specific variables

generally outperform bond market variables for 30 industry portfolios whereas the

650 The Journal of Financial Research