Drawdown By Chuck LeBeau
Drawdown is a topic that is seldom discussed in depth because everyone assumes that its meaning is obvious and that they understand how it should be used in evaluating or capitalizing a trading system. Of all the statistics used in evaluating a trading system's performance I have found that Drawdown, however it might be defined, is a very unreliable statistic at best.
First lets look at how Drawdown is commonly calculated and expressed. This varies widely throughout the industry and even varies within Omega products. For example in TradeStation and SuperCharts, MaxDrawdown is calculated from a position's entry point and not from a position's peak. If we start a trade, which at some point achieves an open profit of $1,000, and then later close it out at a loss of $500, TradeStation and SuperCharts will record a Drawdown of only $500. However, Omega's Portfolio Maximizer would look at the same data and record a Drawdown of $1,500 which is more in line with general industry practice and is the method by which the CFTC and NFA require Commodity Trading Advisors and Pool Operators to calculate Drawdown.
At first glance one might assume that the CFTC/NFA calculation is fairer, or at least more conservative, but that isn't necessarily the case because the regulators require that drawdown be expressed in percentage numbers rather than in dollars. Ask a CTA for their maximum drawdown and you will receive an answer like "35%" not XXX dollars. This makes the size of the drawdown as much dependent on the size of the account at its peak as it is on the size of the decline in equity. A $10,000 drawdown from an equity peak of $100,000 is 10% and the same $10,000 drawdown from an equity peak of $200,000 would be only 5%. Using this formula it would also mean that a trader who never had a loss would still report a drawdown because the giveback of open profits is included as "Drawdown". Omega's Portfolio Maximizer uses this method to calculate Drawdown and it differs substantially from the way that TradeStation and SuperCharts calculate Drawdown.
Even if we assume that we understand which specific Drawdown calculation we are looking at and know how it was calculated, the meaning and application of this statistic are still very suspect. Let's use the trade by trade report of three systems as an example.
System "A" begins trading and has the following closed out trades (we only need a very small sample to illustrate our point):
Loss of $500 Loss of $250 Gain of $2,000 Loss of $500 Gain of $2,500 Gain of $1,000
Using the TradeStation formula this system would have a MaxDrawdown of $750. Keep in mind that if the first trade started out as a $1500 open profit at some point before it was closed out at a $500 loss, the Portfolio Maximizer formula would have recorded the drawdown on this trade as $2,000 not $500. However, lets keep our examples as simple as possible and record the MaxDrawdown for system "A" as $750 (the results of the first two losses, $500 plus $250) as it would show in TradeStation.
Now lets look at the trades for system "B":
Loss of $500 Gain of $2,000 Gain of $2,500 Loss of $500 Gain of $1,000 Loss of $250
Using the TradeStation formula this system would have a MaxDrawdown of $500.
As our final example, lets look at system "C":
Gain of $1,000 Gain of $2,500 Gain of $2,000 Loss of $500 Loss of $500 Loss of $250
Using the TradeStation formula this system would have a MaxDrawdown of $1250.
All three systems made $4,250 and had 50% winning trades. The size of the average winners and losers was identical, as was the Profit Factor (Gross Profits of $5500 divided by Gross Losses of $1250). However, the MaxDrawdown, Account Size Required and Return on Account would all vary substantially.
Which system is best? Which system has the best risk to reward ratio? How should each system be capitalized? Which system is most likely to have the smallest drawdown in the future? In our opinion A, B, and C are all the same system and there is no way to differentiate between them.
Our sample of trades is purposely small but it wouldn't matter much if it were 300 trades instead of only six. In our opinion the sequence of trades in the future will be random and the only valid way to estimate possible drawdowns would be to scramble a large sample of trades and redistribute them randomly. The result might then be expressed as something like this: "Based on a starting capital of $100,000 there is a 28% probability of a drawdown of 50% given 10,000 trials." This drawdown study would require a program that would accept the trade by trade output and then perform a "Monte Carlo" simulation that would redistribute the trades in a different sequence over many trials. Even then our real time experience with drawdowns will be impossible to quantify with any accuracy. The best we can do is to find a range of probabilities and hope that our actual experience falls somewhere within that range.
We know of at least one program that was developed to do this type of drawdown simulation. It is not currently offered for sale but if we can persuade the developer to offer it to our members at a reasonable price we will make it available. There may be other programs that can do this simulation as part of a bigger package but they are generally quite a bit more expensive than the specific software we have in mind.
When building our trading systems we assume that our members would prefer systems that have relatively small historical drawdowns. However we believe that the percentage of winning trades and the size of the winners vs. losers are a much better indication of what to expect in the future than the historical Drawdown statistics viewed out of context. This is one of the reasons you will find that our systems generally emphasize a high winning percentage. In our experience a high winning percentage is the factor that most helps us to control drawdowns.
To summarize, we can not afford to ignore data related to Drawdowns, but we must be very careful how much we rely on this information. As we have attempted to illustrate in this Bulletin, historical Drawdown data tells us very little about what to expect in the future. We suggest that when you look at this data you consider it for the limited value it might have; but don't ever count on it. Of all the performance data we might review, Drawdown and the data calculated from it (Account Size and Return on Account) appear to be the least reliable.
Reprinted with permission by Chuck LeBeau at www.Traderclub.com
