Our backtesting is straightforward:

• Our database that contains a significant amount of information about each listed stock in the US and Canada as well as data about the economy including extensive financial indicators going back on a daily basis many years.
• The database is organized such that only information that was publicly available at that particular moment in time is used in any calculations. For example: a company's earnings for the year start out as a forecast, sometime later that forecast get revised, the company announces their earnings in a press release, then comes the SEC filings, after which there can be revisions as errors in past accounting are discovered, mergers take place, or lines of business are shut down. Therefore the question: "What were GM's earnings for 2001?" depends entirely upon when you asked the question. Every company is included to avoid a survivorship bias.
• Our backtesting works as follows:
• Starting on, say, January 2, 1990, based on what the public knew on that date, what stocks would the model buy?
• The results of that buy order are recorded.
• We advance the clock. The information we know is now different. What stocks in our portfolio does the model want to sell? With those proceeds what stocks does the model want to buy?
• This process continues to the present (or to some other date in the past that we might select).
• We analyze the results paying particular attention to tradability as well as sector, stock and capitalization concentrations.

Our model is not based on minute by minute ticks and there are minimum liquidity requirements for each of our models. Therefore the two most common reasons for the failure of backtests to be duplicated in live trading are not issues for us.

If you want to see how our live trading compares with our backtested results, use the link to the left.