17/04/2020

A mathematical model is believed to have predicted stock market crashes before Covid-19.

Foto: Agência Brasil.

The uncertainty brought about by the new coronavirus has radically affected global markets. In March, the Brazilian stock exchange triggered the "circuit breaker" six times, a mechanism that halts trading, and the Ibovespa fell 30%, the largest monthly drop in the last 22 years. Although the effects of the pandemic have a direct impact on the economic downturn, visiting researchers at IMPA, Ana Cascon from Brazil and William Shadwick from Canada, stated that large declines were already predicted in December of last year.

In “Accidents Waiting to Happen – The US Equity Market,” published in January, financial market experts warned that the S&P 500, an index that lists the largest (by value) stocks listed on American exchanges, could fall 21% in a 20-day period in the following months. Two months later, with the worsening of the pandemic, the S&P 500 had its worst 20-day decline since the 1929 stock market crash, registering a 23% loss on March 20.

“All of our forecasts were based on daily price data up to the end of December 2019. This means that the possibility of US stock markets suffering disastrous falls was already present and identifiable before any impact of the coronavirus on the markets. The probability of suffering large losses has increased and it seems unlikely that the crisis is over,” they state.

Ana Cascon e William Shadwick

Mathematicians also claim to have concluded that the probability of a loss of more than 13% in the S&P, as happened at the end of 2018, was once every eight months. With the increasing impact of the coronavirus on the economy, updated data from March indicate that these losses may be recorded in a shorter period, every 100 days, representing once every five months.

Ana and William have been developing geometry-based statistical tools since 2001 to predict major stock market crashes. Based on these market risk analyses, they can provide precise measures of downside exposure over daily, weekly, bi-weekly, and monthly time horizons. With these tools, the pair says they can give early warnings of asset price bubbles and anti-bubbles, with accurate estimates of potential correction levels. The current projection for the S&P 500 is a drop to the 1300-point level in the coming months.

“We use new tools, based on the perspective that movements in finance are reflections of what has already happened. We use methods based on our research in Probability Geometry, which require neither calibration nor hypotheses. As we are in a very turbulent period, in which new events bring new impacts, this gives greater agility to the results we find,” explains Ana.

Researchers use up to three years of data on fluctuations in the financial market to provide timely answers. “But it’s like a weather forecast. We can predict a 90% chance of rain in Rio, but we can’t say if it will actually rain,” warns the Canadian.

Recently, the pair used the Omega Metrics® suite of “macrostatistical” indicators to analyze the scenario preceding the Great Depression of 1929, formulating “predictions” for the Dow Jones Index (DJI), had the tools been available at the time. “It would have been possible to identify an unstable expansion emerging in the third quarter of 1924, indicating that an asset price bubble was forming. We would have tracked the growing difference between the DJI and its correction level.”

The simulations also pointed to a risk warning in August 1929. “Although there hadn't been major 20-day drops – the temporal parameter for losses used by the study – since June 1921, by the end of August 1929, it would have been possible to identify a one in seven chance of exceeding that loss. The average drop would have been 24% – worse than what happened in the 1907 panic and, in fact, a greater loss than had ever been recorded in the US stock market. Two months later, the market crashed,” they comment.