Market sentiment and its predictive abilities in the stock market : empirical study of leading indicators derived from market sentiment

2021

The purpose of this thesis is to investigate the relationship between multiple possible leading

indicators and the stock market (S&P500 index). A leading indicator can be defined as a piece

of data that corresponds to the future movements in a variable of interest. Thus, considering

the information they contain, one may be able to anticipate the movement of the stock market.

The indicators to examine are representations of market sentiment, reflecting the risk tolerance

of investors in the financial markets. They have been determined based on existing literature,

in addition to reflections of the authors of this thesis. For example, the Put/Call ratio (Pan and

Poteshman, 2003) and the Gold/Platinum ratio (Huang and Kilic, 2019) has been shown to

have predictive abilities towards the stock market. Previous research has mainly focused on

the potential leading indicators individually. To explore their abilities combined is therefore

of interest.

The results appear to favorize the inclusion of multiple indicators in this analysis. However, it

shows conflicting evidence for the predictive abilities of the indicators toward the stock index.

Parts of the analysis appear to show predictive capabilities, while others are less conclusive,

which complicates the utilization of the results. The patterns seem to change as different

periods are examined, but when changes in the economic environment are considered many

of these observations seem reasonable.

A vector autoregressive model (VAR-model) is facilitating the use of both the present and past

notations of the indicators in a system. Multiple models were created, one where the whole

sample period was included, in addition to several subsets to display the potential changes that

occur. The model including the whole sample period (2005-2020) is reporting an adjusted R2

of 12,0%, where the subsets range from 18,1% to 28,4%, suggesting that the importance of

the indicators varies through time, as different pattern emerges.

The results from the VAR-model are evaluated using granger causality test and impulse

response analysis. The indicators appear to be granger causal throughout the analysis, with

only one exemption in one of the subsets. In the impulsive response functions, a theoretical

shock is performed on the indicators. It illustrates that whenever there is an increase in

indicators like the Put/Call ratio, the U.S. Dollar index (the relative strength of the dollar) and

the credit spread (Baa-Aaa), the stock index appears to consistently react negatively to this.

The Gold/Platinum and the Volume Weighted Moving Average looks to result in rather

inconsistent reactions. The decomposition of the variance shows that most of the variance

comes from the shocks in the stock index itself, which is the recurring observation throughout

the subsets. This is to be expected in the beginning of the simulated period. Later, more of the

variance is explained by the other indicators. Nonetheless, the increase should have been larger

to claim it supports the hypothesis of the predictive abilities.