Sentiment indices and their forecasting ability

• Published date: 01 July 2019
• Author: Frank J. Fabozzi,David A. Mascio
• Date: 01 July 2019
• DOI: http://doi.org/10.1002/for.2571

Received: 15 June 2018 Revised: 26 November 2018 Accepted: 8 January 2019

DOI: 10.1002/for.2571

RESEARCH ARTICLE

Sentiment indices and their forecasting ability

David A. Mascio1Frank J. Fabozzi2

1Della Parola Capital Management, Fort

Collins, Colorado, United States

2EDHEC Business School, Nice, France

Correspondence

Frank J.Fabozzi, EDHEC Business School,

57 S. Main Street, Doylestown, PA18901.

Email: fabozzi321@aol.com

Abstract

The success of any timing strategy depends on the accuracy of market forecasts.

In this paper, we test five indices to forecast the 1-month-ahead performance of

the S&P 500 Index. These indices reflect investor sentiment, current business

conditions, economic policy uncertainty, and market dislocation information.

Each model is used in a logistic regression analysis to predict the 1-month-ahead

market direction, and the forecasts are used to adjust the portfolio's beta. Beta

optimization refers to a strategy designed to create a portfolio beta of 1.0 when

the market is expected to go up, and a beta of −1.0whenabearmarketis

expected. Successful application of this strategy generates returns that are con-

sistent with a call option or an option straddle position; that is, positive returns

are generated in both up and down markets. Analysis reveals that the models'

forecasts have discriminatory power in identifying substantial market move-

ments, particularly during the bursting of the tech bubble and the financial

crisis. Four of the five forecast models tested outperform the benchmark index.

KEYWORDS

beta optimization, market timing strategy, sentiment index, stock market forecasting

1INTRODUCTION

“Beta optimization” is a market timing strategy designed

to alter the beta of a portfolio as equity market expectations

change over time. The traditional market timing strategy

attempts to generate excess returns by switching between

common stocks and bonds or cash equivalents as the stock

market fluctuates. If an investor can consistently iden-

tify the turning points in the market, the timing strategy

can substantially enhance investment performance. Origi-

nally, the market timing objective wasto be long common

stocks (100%) during bull markets and long cash equiv-

alents (100%) during bear markets. Timers make their

portfolios more or less sensitive to the market by switch-

ing from stocks (beta approximately 1.0) to bonds or cash

(beta approximately zero) and back as their outlook for the

market changes.

Portfolio betas can be increased to 1.0 through the use

of leverage or can be made negative by shorting stocks or

index futures. The beta optimization market timing objec-

tive is to create a portfolio with a high beta when the

market is expected to go up, and a beta of zero when the

market is expected to be neutral. When a bear market is

expected, a conservative investor will use cash or bonds

to create a zero-beta portfolio, whereas a more aggressive

manager will use short positions to create a negative beta

portfolio. Successful application of this strategy generates

returns that are consistent with a long call option posi-

tion when using cash or bonds; a look-back option straddle

payoff is realized when short equity positions are used to

create a negative beta. In either case, positive returns are

generated in both up and down markets.

Research into market timing has a long history.

For example,Treynor and Mazuy (1966) point out that suc-

Journal of Forecasting. 2019;38:257–276. wileyonlinelibrary.com/journal/for © 2019 John Wiley & Sons, Ltd. 257

258 MASCIO AND FABOZZI

cessful market timing managers will shift the composition

of their portfolios between more and less volatile securities

as their outlook for the markets change over time. Merton

(1981) points out that a market timer or “macro-forecaster”

tries to predict whether stocks (bonds) will outperform

bonds (stocks), and that the returns generated by success-

ful market timers' returns are “virtually indistinguishable”

from successful option strategies. The resultant payoffs for

successful market timers resemble that of a call option;

funds are in stocks during market increases and in bonds

or cash during stock market declines. If the manager has

the ability to use short selling, a look-back option payoff

can be generated.

The success of any timing strategy depends on successful

market forecasting, and market participants are contin-

ually experimenting with modeling changes in financial

markets. While a brief overview of forecasting is presented

below, in this paper five recently developed indices are

used to forecast the direction of the market (the S&P 500)

in the upcoming month. The S&P 500 Index (SPX) with a

beta of 1.0 is considered the market portfolio.

In our tests, if a bull market is forecast, a long position in

the SPX results in a beta of 1.0. Obviously, in practice the

beta of the portfolio can be increased through the use of

leverage or margin. If a bear market is expected, a zerobeta

is achieved by moving from the SPX into cash equivalents,

orabetaof−1.0 is generated by going short the SPX. Both

strategies are examined and compared.

We examine the efficacy of the beta optimization strat-

egy using three types of market forecasting models, and

the market timing results of these models are compared

to a benchmark index by Baker and Wurgler (2004).1An

investor sentiment index developed by Baker and Wurgler

(2007) represents the first type of model. This model com-

bines six proxies for sentiment to generate a “sentiment”

index.

The second two models are the ADS index from the

Philadelphia Federal Reserve Bank, developed by Aruoba,

Diebold, and Scotti (2009), and the improved investor sen-

timent index, developed by Huang, Jiang, Tu, and Zhou

(2015). The ADS model combines quarterly, monthly, and

weekly macroeconomic information to develop an outlook

for the overall economy,while the Huang et al. model was

created to forecast aggregate stock market performance

and is a proxy for macroeconomic variables.

The final two models are the Gilchrist and Zakrajsek

(2012) GZ Spread Index and the Jurado,Ludvigson, and Ng

(2015) Financial Uncertainty Index. The GZ Spread Index

is a corporate fixed-income credit spread model, and the

financial uncertainty index is an independent time-varying

1An online supplement (Supporting Information) provides a detailed

description of each model, and is available upon request.

econometric model. Both are created to predict the overall

financial market risk.

The stock market forecasting ability of these five empiri-

cal forecasting models are assessed based on the following

criteria. First, the number of monthly market movements

(both up and down) are correctly forecasted. Second,

the effectiveness of each model in predicting the major

down-market movements, the so-called “black swan”

events (BSEs) that can devastate portfolios, is assessed.

Avoiding the worst months can enhance portfolio returns

substantially: Some models may be superior in forecast-

ing severely down markets. Third, each model's predictive

power based on its statistical significance is evaluated.

Finally, all forecasts are compared to two perfect monthly

timing strategies (or perfect forecasting strategies).

While the SPX represents a one-asset portfolio, the pro-

cedures outlined here can be useful for both individual

investors and asset managers. Individual investors can

switch between equity mutual funds and bond or money

market funds, while active portfolio managers can adjust

the betas of their portfolios by increasing (decreasing) port-

folio betas through stock selection and switching to bonds

or money market instruments or to short positions as their

outlook for the market changes.

Many studies provide evidence that both support and

refute an investors' ability to successfully time the market.

Research since the turn of the century that supports the

ability to time markets include Bollen and Busse (2001),

Fung and Hsieh (2001), Lam and Li (2004), Jiang, Yao,and

Yu (2007), Avramov and Wermers (2006), Cremers, Mar-

tijn, and Petajisto (2009), Kacperczyk, VanNieuwerburgh,

and Veldkamp (2014). In contrast, Hubner (2010), Jagan-

nathan and Korajczyk (2015), Fisher and Statman (2003),

and Brown and Cliff (2004) conclude that market timing

may not be a worthwhile endeavor. In this paper, we pro-

vide support for market timing efforts. We demonstrate

through our beta optimization methodology that investor

sentiment does in fact predict future stock returns.

The paper proceeds as follows. The data and method-

ology are presented in Section 2. Empirical results are

reported in Section 3. The effectiveness of the models' abil-

ity to forecast BSEs and the Financial Crisis are described

in Section 5. Concluding remarks are in Section 6.

2DATA AND METHODOLOGY

2.1 Data

We employ five models to forecast the movement of the

stock market 1 month ahead. The models arethe Baker and

Wurgler (2007) Sentiment Index (SENT), the Huang et al.

(2015) Improved Sentiment Index (SI), the Aruoba et al.