Transform asset price time series into stationary form apply Discrete Fourier Transform to eliminate noise and identify/model repeating patterns to exploit and backtest.
The project was inspired by the following academic journals:
- This digital signal processing technique is used to de-noise the data. Discrete Fourier Transform Splits time series into a set of waves. These waves can be summed up to construct the original time series again!
- However, objective is to remove noise from original time series.
- Therefore, the concept is that, we only select a few set of waves with most amplitude -> and only use these to reconstruct the time series. the resulting time series will only have the most prominent characteristics of the original.
- Transform original time series to find a repeating pattern/anomaly.
- Check for stationarity using Augmented Dickey Fuller test
- If not stationary, then detrend the data to remove seasonal factors to make it stationary
- Eliminate noise in the pattern to better model it, this is where digital signal processing come in
- Apply Discrete Fourier Transform to split the time series into different wave functions
- Only take the most prevalent wave functions and rebuild the time series. This would theoretically remove noise within the time series.
- Try to take advantage of this repeating pattern in the market
- Incorporate risk management measures in the strategy
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent