Interest rate level and stock return predictability

• DOI: http://doi.org/10.1002/rfe.1059
• Published date: 01 October 2019
• Author: Yaojie Zhang,Dengshi Huang,Yongsheng Yi,Feng Ma
• Date: 01 October 2019

506

|

wileyonlinelibrary.com/journal/rfe Rev Financ Econ. 2019;37:506–522.

© 2019 University of New Orleans

1

|

INTRODUCTION

Forecasting stock market returns is a charming topic, appealing to both researchers and investors. Researchers focus on the

operations of the stock market and its efficiency, whereas individual investors utilize the predictability of stock returns to

assist in their decision- making and to initiate portfolio strategies. The pursuit of accuracy in stock return predictions has a

long history. Considering in- sample estimates, an impressive body of existing literature documents the predictability of stock

returns (Ang & Bekaert, 2007; Lewellen, 2004; Stambaugh, 1986; Welch & Goyal, 2007), but this work achieves a poor out- of-

sample performance that is even worse than the performance of a benchmark model assuming constant returns. The forecasts

based on single economic fundamentals proposed by Welch and Goyal (2007) can hardly beat the historical average bench-

mark.1 However, recent studies have made great progress, while imposing constraints on single- predictor predictive models

(Campbell & Thompson, 2008; Pettenuzzo, Timmermann, & Valkanov, 2014). Furthermore, Rapach, Strauss, and Zhou (2010)

find that the multivariate strategies of combining individual forecasts can over time statistically and economically outperform

the benchmark.

It has been documented that macroeconomic factors have an effect on the variation of aggregate stock returns (Flannery

& Protopapadakis, 2002), but nearly all univariate predictive regressions based on single predictors cannot accurately predict

stock returns. Regarding this issue, considering that the market is not able to rationally reflect fundamentals, Summers (1986)

argues that the efficiency of financial markets cannot be supported. In addition, his work further suggests that there are large

valuation errors in speculative markets. We can also learn from the research by Scheinkman and Xiong (2003) that stock prices

Received: 23 July 2018

|

Revised: 8 January 2019

|

Accepted: 12 February 2019

DOI: 10.1002/rfe.1059

ORIGINAL ARTICLE

Interest rate level and stock return predictability

YongshengYi

|

FengMa

|

DengshiHuang

|

YaojieZhang

School of Economics and

Management,Southwest Jiaotong

University, Chengdu, China

Correspondence

Feng Ma, School of Economics and

Management, Southwest Jiaotong

University, Chengdu, China.

Email: mafeng2016@swjtu.edu.cn

Funding information

Natural Science Foundation of China,

Grant/Award Number: 71671145 and

71701170; Humanities and Social Science

Fund of the Ministry of Education,

Grant/Award Number: 17YJC790105;

Fundamental Research Funds for the

Central Universities, Grant/Award Number:

2682017WCX01 and 2682018WXTD05

Abstract

We employ a novel interest rate- determined model- switching strategy to forecast

stock returns and find persistent predictive ability among a number of economic

fundamentals. This strategy switches predictive models based on whether a real- time

interest rate is higher than the mean level interest rate of a look- back period. The

robustly better predictive ability of the new strategy relative to the original OLS re-

gressions suggests that stock returns react more rationally to the variation in eco-

nomic fundamentals if the contemporaneous interest rates are high. This pattern is

consistent with prior literature demonstrating that a high interest rate attenuates spec-

ulative demand (Keynes, The general theory of employment, interest, and money,

Harcourt Brace, London, 1936), while speculative trading drives stock prices away

from their valuation founded on economic fundamentals (Scheinkman & Xiong,

Journal of Political Economy, 111, 1183, 2003).

JEL CLASSIFICATION

C22, G11, G12, G14

KEYWORDS

forecasting performance, interest rate level, model switching, predictive regression, stock return

|

507

YI etal.

contain a large non- fundament al component and that speculative trading among agents creates bubbles in the stock market, thus

driving the prices away from their valuations founded on economic fundamentals. This proposition is also supported by earlier

studies suggesting that the stock price contains two components: one component implied by the efficient market model and the

other created from a speculative bubble (Blanchard & Watson, 1982). With these propositions in mind, we consider whether

economic fundamentals can better predict stock returns when the speculative excitement can be recognized ahead of time. It

is well known that interest rates influence the aggregate movement of financial markets and, more importantly, that interest

rates also affect the investment behavior of participants in financial markets. We learn from Keynes (1936) who claims that

the speculative demand for money is negatively associated with interest rates. We interpret this proposition as indicating that

a high interest rate may be able to attenuate the excitement of speculation. In addition, Hong, Scheinkman, and Xiong (2006)

demonstrate that the price is often biased upward during the formation of speculative bubbles because prices only reflect the

beliefs of optimistic investors, whereas the pessimistic beliefs are suppressed by short- sales constraints. Thus, a high interest

rate may prevent investors from conducting speculative investments in the stock market because of a high opportunity cost or a

leverage- related cost, while at the same time, the short- sale constraint also mitigates the investor's speculative demand for less

risky assets with high interest rates. Under this condition, a high interest rate may result in a relatively rational operation of

the stock market. In contrast, a low interest rate may stimulate the demand for speculative investment in the stock market and

yield biased stock prices or even speculative bubbles, which finally leads to valuation errors in the stock market. An incorrect

reflection of economic fundamentals by stock prices thus can to some extent be attributed to low interest rate- induced specu-

lative trading behavior. Moreover, Fang and Bessler (2017) also find evidence that interest rates contribute to the predictability

of stock returns. Naturally, we wonder whether economic fundamentals can better predict stock returns over the periods when

the corresponding rates of interest are high. With this motivation, using economic fundamentals and considering the level of

interest rates simultaneously, we employ a new strategy to forecast stock returns.

Our new forecasting strategy builds on an interest rate- determined model- switching (IRDMS) mechanism. Specifically, we

propose that considering the relationship between the interest rate level and the operation of the stock market, when the con-

temporaneous interest rates are high, economic fundamentals can better predict stock returns. To conduct the model- switching

strategy, we propose several criteria to judge whether the real- time interest rates are above a normal level. When a real- time

interest rate is higher than the mean value of rates over several preceding months, the corresponding rate can be classified as

a high level, and we forecast the stock return through regressions. Alternatively, when the stock market is operating irratio-

nally, we think the subsequent stock returns are difficult to forecast, and we instead use historical averages for the forecast.

Considering the availability of interest rate data for our long- sample empirical analyses, we employ the secondary market rates

of the 3- mont h U.S. treasury bills (TBL) as the proxy for short- ter m interest rates. At the same time, we also employ long- term

government bond yields (LTY) as an alternative proxy for robustness. Overall, regarding the look- back periods of 3 months,

6 months, 12 months, 18 months, and 24 months, the IRDMS strategies related to TBL are IRDMS- TBL(3), IRDMS- TBL(6),

IRDMS- TBL(12), IRDMS- TBL(18), and IRDMS- TBL(24), respectively, while the strategies related to LTY are IRDMS-

LTY(3), IRDMS- LTY(6), IRDMS- LTY(12), IRDMS- LTY(18), and IRDMS- LTY(24), respectively.

To validate the efficiency of our IRDMS strategy, we employ univariate predictive regressions to generate out- of- sample

forecasts. The 12 predictors utilized in this study are the variables suggested by Welch and Goyal (2007), and they cover a

sample period spanning January 1927 to December 2016; the out- of- sample period spans from January 1957 to December

2016. We follow Campbell and Thompson (2008) to employ the out- of- sample R2 statistic (

R2

OS

) to evaluate the forecasting

performance. A positive

R2

OS

indicates that a predictive model of interest performs better than the historical average benchmark.

We also employ Clark and West's (2007) (CW) statistic to evaluate the significance of the forecasting performance. Consistent

with the results from Welch and Goyal (2007), who demonstrate that single predictors can hardly accurately forecast stock

returns, the findings from our empirical tests suggest that 10 of 12 original predictive models generate negative

R2

OSs

. After

implementing the IRDMS strategy, most of these univariate predictive regressions improve their performance and a number of

the IRDMS- based models even generate significant positive

R2

OSs

. In terms of the five TBL- related criteria, after implementing

the IRDMS strategy, six predictors, i.e., DP, DY, TBL, LTY, LTR, and INFL consistently generate positive

R2

OSs

. Among all 12

predictors, TBL and LTY are the best two predictors. In particular, TBL achieves the greatest improvement with the application

of IRDMS- TBL(24), and the corresponding

R2

OS

improves from 0.08% to 1.298%, significant at the 1% level. Meanwhile, the

greatest improvement for LTY is achieved through IRDMS- TBL(18), and the corresponding

R2

OS

increases from −1.027% to

1.711%, which is also significant at the 1% level. For the five LTY- related criteria, the best performance is still achieved by

TBL or LTY. Notably, after implementing IRDMS- LTY(6), the

R2

OS

for TBL rises to 1.129%, and for LTY, it rises to 0.827%.

Investors usually pay more attention to the economic benefit of return forecasts, so we also follow Welch and Goyal

(2007), Rapach et al. (2010), and Ferreira and Santa- Clara (2011) in calculating the certainty equivalent return (CER) of

a mean- variance investor who allocates her assets between risk- free assets and the S&P500 index. From the results, we