10 – M4 L3a 09 Smoothing V2

Financial data is noisy, and sometimes the data we’re working with is sparse. For instance, it might have many missing values. We can apply smoothing techniques across the time dimension to help make the factor more robust to noise and sparse data. Fundamental data is also sparse in nature, since it is only updated in the US, for example, once every three months. If we wish to use in a higher frequency, such as daily frequency, we should copy the most recent data over to each new day until the next new data point arrives three months later. We could also apply smoothing so that the daily data points incorporate some weighted averaging of not only the most recent update, but also the value from previous quarters. To make our Alpha factor more robust to noise and sparse data, we can apply a rolling window average. A rolling window is a kind of smoothing technique. A variation of a rolling window average is a weighted average, where the most recent Alpha may be given more weight and prior Alpha values are included but given less weight. This is called linear decay. For example, if we choose a window length of two and took the weighted average of the most recent Alpha values and the prior days Alpha values, we could give the newest value a weight of two and the earlier value a weight of one. In general, for a window length of T, we would give the most recent Alpha value a weight of T. Then for the earlier alpha values, we’d give them smaller weights. This technique can be very effective. In fact, I’ve seen several cases where the application of a smoothing operator both increase the Sharpe ratio and decrease the turnover. By the way, we’ll cover the Sharpe ratio and a proxy for measuring turnover later in this lesson. These are both important evaluation metrics that help us find out if an Alpha factor that we create has the potential to enhance a portfolio’s performance.

%d 블로거가 이것을 좋아합니다: