Mechanical Trading Systems: Pairing Trader Psychology with Technical Analysis

PITFALLS OF VALUE AT RISK

Although value at risk is a valuable tool, it does not provide a comprehensive solution to the problem of managing price risk. This is because VaR does not address how much we could lose during a given holding period, only the maximum that we are likely to lose. For example, while VaR defines an excessive loss for our particular trading account as $30 million and tells us that we have a 5 percent chance of enduring such a loss over the next 24 hours, it says nothing about how severely any particular daily loss will exceed this $30 million threshold.

Another flaw in VaR models is that they assume serial independence: whatever happened today has no impact on tomorrow's trading. Because all of the trading systems discussed in this text owed their success to the market's propensity to either trend or revert to the mean, serial independence is a flawed assumption.

The assumption of serial independence leads to VaR's inability to account for excessive event clusterings. For example, the fact that we exceeded our daily VaR threshold yesterday says nothing about the likelihood of VaR being exceeded again today. In fact, excessive event clusterings can be one of the distinguishing traits of a strongly trending market. This is perhaps best exemplified by the 1995 Mexican peso crisis, during which the market experienced 9 days beyond 20 standard deviations from the mean.

A subtler problem relating to cumulative losses is that VaR only attempts to predict the likelihood ...