If possible, we’d like to generate trading signals that are robust to outlying data points. For example, imagine you have an outlier in the closing price time series, a huge positive spike. If you’re using a momentum based strategy and your strategies buy or sell decision is based solely on individual closing price values, then the single value might induce a buy decision. If instead you take the moving average of closing prices averaged with a fixed duration rolling time window as the input to your signal, it will average out that single extreme datum and reduce the effect of individual outliers on buy or sell decisions. The trade-off here is that your signal actions might be generated with a slight delay relative to the stock price movement. The duration of this delay will depend on the window size you use. You can take an analogous approach for portfolio level strategies, or strategies that are based on the movements of an entire portfolio of stocks,. You can average out the extreme movements of individual stocks by basing your buy or sell decision on the accumulated movements of many stocks, or even an entire sector. This will also reduce the effect of outliers on your signal. If you are looking for alternative solutions, there have been ongoing efforts from within the industry to incorporate Bayesian methods, or machine-learning, into outlier detection. This is outside the scope of this lesson, so, we will leave the subject to you to explore.