• Performance
• Services
• Company
• Learn About
• My Attain

Outlier Trades: Friend or Foe.

April 26, 2004

As you just read above, last week was an unbelievably great week for several day trading systems. No sooner had we finished last week's newsletter warning investors of jumping on the next HOT system; than several of those HOT systems posted huge winners. More than $5,000 +a piece for Helix SP and R-Mesa SP per contract on a single trade Tuesday were the highlights.

The $5,100+ R-Mesa 5 winning trade took that system from down ~11% for the year to UP ~6% in the span of a few hours. (Percentages based on the developer's recommended initial balance of $30,000) Such a dramatic turn around in just a single trade is literally unbelievable, and brings to the forefront the issue of whether such outlier trades are a good thing for a trading system.

I can hear people asking how such a trade could be anything but a good thing. Who doesn't like big winners? But the possibility that such a trade is a one time anomaly not likely to be repeated is what makes such outliers problematic. If an investor decides to follow a certain trading system based on that system's statistics, but those statistics are skewed upwards by an outlier trade or two - there is the very real possibility that the hypothetical results are overstated somewhat.

So can the investor choosing to ignore last week's newsletter and get involved with one of these systems expect such trades in the future? That is the hundred thousand dollar question, of course; and the quick answer is no. Investors should not expect to see such large winning trades, as they are statistically few and far between. We call such trades outliers.

An outlier is a statistical term. It refers to observations in a distribution of data that deviate so much from the other observations as to arouse suspicions that it was generated by a different mechanism. Statisticians hate outliers. They make their data poor fits onto their models, and are therefore normally discarded so as to get more accurate statistics.

But how should the system investor view outliers? As friend or foe. Should we remove them out of backtested results as the statisticians would have us do, throwing out the highest and lowest values to give a possibly more accurate representation of future performance? Or can we accept that outlier trades are not statistical anomalies, but real trades not only destined to happen, but indeed planned for in a system's logic.

The scary part for investors thinking they may want the outliers removed from backtesting is that trading systems do rely heavily on outliers. The table below shows a cross section of systems with the number of outliers they have had over the past xx years, the dollar amount of profit attributed to those outliers, and the percentage of the total net profit those outliers represent. You will see that the systems in this example have relied rather heavily on outliers, with the total profit of each reliant on outliers by double digit figures.

But is this a bad thing? Many systems are designed to capture such outliers, giving them more of a long option profile. The long option profile results in a distribution of returns tightly bunched around very small losing trades, with a small number of very large outlier winning trades. A short option profile, in contrast, has many small winning trades, but a few devastatingly large losing trades. Most system developers are no dummies - they don't want a short option profile which would result in eventual devastation.

Most system developers don't set out to create a system dependent on outliers, but do so by following the simple tenet of "let winners run, while cutting losers short". This philosophy results in an abnormal distribution of trade results, where there are more "extreme" results sitting way out on the right hand of the curve than could be expected by a normal bell shaped curve. Long term, multi-market systems rely more on the "Let Profits Run" mantra than their day trading counterparts, for the main reason that the day trading systems have just a single day to let their profits run.

There are such things as negative outliers, but systems avoid these like the plague. It's just not worth it to risk a very large amount on many trades for the higher probability of many small winners. If you see even a small number of negative outliers in backtested results, be worried, as such a system has most likely adopted the short option profile.

So if you see some positive outliers in your actual trading or backtested results, it should not be cause for alarm. One caveat to this can be if several outliers occur across a single time period. For example, many Nasdaq trading systems look very good in backtesting, showing hundreds of thousands of dollars in profits. Looking more closely at the statistics, however, one will find that the grand majority of such profits were attained in the NASDAQ's hay day of 100 point moves in late 1999/early 2000.

By nearly all accounts, the above mentioned time period was a speculative bubble and collapse the likes of which may never be seen again, and putting too much faith in results which owe the grand majority of their profits to that time period can lead to disappointment in the current environment where the Nasdaq range is just 20% of what it was.

So beware the abnormal conditions outlier in backtesting, but welcome with open arms the real-time outlier which pulls your system into the black for the year.