Thinking about how to deal with Outliers is an important part of signal research. Let me give you a couple examples. Sometimes, the prices of stocks that aren’t traded very much undergo dramatic changes when they are actually traded. As an example, let’s look at some small market capitalization stocks traded on the Hong Kong Stock Exchange. One example is Easyknit International Holdings Limited which has the ticker symbol 1218.HK. I’m using Easyknit as an example of a thinly or infrequently traded stock, but what does it mean to be thinly traded. Well, one way to quantify how much a stock is traded is to look at its typical volume. This is the number of shares that were traded during a given time period. However, the number of shares traded may not tell you very much since a company can issue arbitrary numbers of shares. A better metric called turnover is the volume traded multiplied by the price per share, in other words, the total amount of money that actually changed hands when the stock was traded during this period. This metric is more comparable across companies since it’s in units of currency. Easyknit’s average volume was only about 10,000 shares at the time we made this video and traded at around 4.8 Hong Kong dollars for a turnover of around 48,000 Hong Kong dollars. For comparison, the turnover of Tencent Holdings, a leading provider of Internet related services in China and one of the most traded stocks on the Hong Kong Stock Exchange, was around five billion Hong Kong dollars. It doesn’t take much trading to move the prices of thinly traded stocks up and down. A single trade will cause a significant change in price, so the prices can fluctuate unpredictably. These fluctuations are real, they are due to market activity, but they are unpredictable. You could avoid exposure to this unpredictability by excluding these stocks from your trading universe. Alternatively, if you want to trade these stocks, it might be a good idea to keep these price fluctuations in your trading data in order to be sure that your strategy performs well despite this unpredictability. This example illustrates why it’s important to be careful which stocks you chose to include in your trading universe in the first place. Including data from periods when the market is behaving in an unusual way can skew your signals. For example, during and after market crashes, stock prices are volatile and can reach extreme values. Because market crashes do not happen often, they are statistically infrequent outlying events. If we calibrate trading signals including data from these periods, the results will be highly skewed and the signals won’t perform optimally on normal trading days, in other words, most of the time. But if we don’t include such periods when calibrating trading signals, and one of these rare events happens, the signal may perform really poorly or even in a toxic manner. There is no golden rule to resolve this, some traders prefer to alleviate these problems in the strategy formulation phase by controlling the sizes of the long and short positions they take and establishing thresholds called stop loss levels at which they will exit positions to prevent further losses. It’s also possible to design training strategies that attempt to take advantage of sudden extreme price movements. People who trade with so-called contrarian strategies seek to identify sudden price movements or dislocations that they think are not justified by real information or events and therefore won’t last, and they trade against such movements. One could try to apply this strategy to data of any level of time resolution, but minute level data are more prone to these sudden movements than daily data or monthly data. This is a risky strategy, it can be difficult to determine whether the price will settle back to where it was, in which case the trader profits, or whether the price will continue to move in the same direction which would not benefit the trader.