On the predictability of stock market bubbles: evidence from LPPLS confidence multi-scale indicators

Abstract

We examine the predictability of positive and negative stock market bubbles via an application of the LPPLS Confidence Multi-scale Indicators to the $S\mathrm{\&}P500$, FTSE and NIKKEI indexes. We find that the LPPLS framework is able to successfully capture, ex-ante, some of the prominent bubbles across different time scales, such as the Black Monday, Dot-com, and Subprime Crisis periods. We then show that measures of short selling activity have robust predictive power over negative bubbles across both short and long time horizons, in line with the previous studies suggesting that short sellers have predictive ability over stock price crash risks. Market liquidity, on the other hand, is found to have robust predictive power over both the negative and positive bubbles, while its predictive power is largely limited to short horizons. Short selling and liquidity are thus identified as two important factors contributing to the LPPLS-based bubble indicators. The evidence overall points to the predictability of stock market bubbles using market-based proxies of trading activity and can be used as a guideline to model and monitor the occurrence of bubble conditions in financial markets.

Keywords:

• Financial bubble indicators
• LPPLS model
• Markov switching
• Predictability
• Short interest

JEL classification:

• C13
• C58
• G14

Acknowledgments

We would like to thank an anonymous referee for many helpful comments. However, any remaining errors are solely ours.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Riza Demirer http://orcid.org/0000-0002-1840-8085

Notes

† A positive (resp. negative) bubble is defined as an upward (resp. downward) accelerating price followed by a crash (resp. rally).

† For further information about the sloppiness of the LPPLS model, we refer the reader to Demos and Sornette (2017) and Filimonov et al. (2017).

† see http://www.er.ethz.ch/financial-crisis-observatory.html

† Given that these indicators are available only for the S&P/500 Index, we restrict testing for the predictive power of short selling activity to this market thus deliberately excluding both FTSE and NIKKEI from the analysis.

† Note that, in addition to the Bai and Perron (2003) test of structural breaks, we also conducted a Bayesian change point analysis as proposed by Barry and Hartigan (1993). In this case, the algorithm assigns posterior probability of a change for each date of the variable under consideration. Using the proposed prior values of Barry and Hartigan (1993), we identified break dates whenever the posterior probability for a change on a specific date was greater than equal to 0.50. Based on this, we were able to identify many more break dates for the S&P500 price-dividend series, and also for the six indicators of bubbles. In comparison to the 5 break dates identified by the Bai and Perron (2003) approach, the corresponding closest break dates detected by the Bayesian change point analysis were: 1981M08, 1987M10, 1996M12, 2001M02, and 2008M06. More importantly though, the break dates from the bubbles indicators preceded those of the S&P500 price-dividend ratio, and in most cases, were immediately before the break date detected for the price-dividend ratio. Complete details of the Bai and Perron (2003) structural break results and that of the Bayesian change point analysis are given in table 8 for the S&P-500 Index.

‡ The model is estimated using the non-smoothed bubble indicators explained in section 3.3 as the smoothed indicators may lead to spurious regressions and inaccurate p-values, as mentioned earlier.

† We thank an anonymous reviewer for this suggestion.