The Valeant Case Study: Could a Fat-Tailed Market Risk System Have Warned of Valeant’s Fall?

Posted by Martin Dimitrov, Client Services Manager, BISAM on Apr 5, 2016 9:00:00 AM
Find me on:

Following news of Valeant Pharmaceutical’s worst ever single day loss on March 15, 2016, BISAM’s client services team was challenged by a fund manager, who asked:  Could BISAM’s Cognity market risk solution have predicted the VaR forecast for Valeant in the weeks and months leading up to and just after news of its sell-off hit the headlines?

The simple answer is: No. But the answer is actually not so simple. In fact, the Cognity market risk system arms its users with the tools essential to understanding and modeling specific risk. So while Cognity cannot predict these events directly, it can certainly help risk and portfolio managers to evaluate and model such risks.

Here is Our Team’s Quantitative Analysis to Show How

Two key and unique qualities of the Cognity platform were used throughout the following analysis:

  1. The Cognity fat-tailed methodology, which models extreme events
  2. The depth of Cognity’s factor modeling, transparency and customization capabilities

We started with a simple experiment running a 99% VaR backtest on Valeant (VRX) to see if the Cognity proprietary methodology would have spotted anything on its own compared to the traditional Normal model. Note: the Normal model used includes volatility clustering (GARCH model).


As you can see, the Cognity model is more realistic as it becomes more conservative at the very early signs of stock price vulnerability. Looking at the 4-week moving average - the Fat-tailed-Normal VaR spread - we observe that it doubles before the first stock price drops in September, from 0.64 in late August to 1.33 in mid-September. Such an increase can be considered an indication of increased turbulence of the returns. Still, this cannot be interpreted as a warning on the upcoming events.


However, look what happens when we blend Cognity's strong quant analytics with qualitative information and incorporate it into factor models:  

In particular we explored the warning made by Mr. Khmelnitsky on possible business malpractices, and chose to incorporate this view into the analysis as of Q3 2014, when his first “sell” signal was announced.

We chose to incorporate this scenario by creating a “malpractice” factor in Cognity:

1. Select companies with similar events:
  • Waste Management (1999)
  • Enron (2001)
  • Tyco International (2001)
  • WorldCom (2002)
  • HealthSouth (2002)
  • Lehman Brothers (2008)
  • Satyam Computer Services (2009)
  • MF Global (2011)


We formed the “de-timed” and “de-marketed” (i.e. “specific”)  returns of the stocks by regressing towards a general market index (S&P 500), and choosing 1.5 years before the first large stock drop, and half a year after events unfolded. Residuals from such regression can be considered to be a time-series with a unified time stamp as general market behavior and market cycles (defined by the time-period) are “cleaned” by the regression. The new time scale is now defined by the stock burst event.


2. Create a “malpractice” factor by running Principle Component Analysis on the de-marketed, de-timed (specific) return series:


The first factor explains close to 40% of the variance of the stocks.
We incorporated it into the analysis by customizing the Equity section of the Global Multi-Asset Class Cognity Model.


Running the backtest with the updated factor model, we see stronger warning indications with VaR jumping to 20% before the October losses.



In Conclusion

The Cognity fat-tailed model applied to model the VaR of Valeant stock prices is indeed more reliable that the Normal approaches. Still, no purely statistical model can predict such “Black Swan” events in the lifetime of a stock. However, in this case we consider the powerful factor modeling capabilities in Cognity - their flexibility and transparency – to be of greater help: users are able to embed qualitative judgment and blend them with purely statistical analysis in order to form realistic expectations for the otherwise unforeseeable future. Such an analysis, typically considered as heavy research, can be executed solely in the advanced suite of modules available (Cognity Research Module) in the Cognity platform. More importantly, such analysis can be executed in a matter of minutes in a smooth workflow, with no coding at all.


Cognity_info__request.pngFinAnalytica's  Cognity® platform, now part of BISAM’s suite of market-leading portfolio analytics, is  a unique, comprehensive, multi-asset class solution for market risk, portfolio construction and investment decision analytics – designed specifically for the buy-side. Visit to request a demo.



Topics: Risk Management

Leave a Comment

If you have any comments or have any questions about the subjects raised in this post, please fill in your details and message below. We'd like to hear from you.