SSA Trend Predictor

Controlling the indicator parameters allows you to adjust the smoothness of the selected trend and control the noise filtering threshold, take into account or suppress the contribution from the components in the signal acting on different time scales. The optimal decomposition of data into signal and noise components is determined by the time horizon of the trading strategy and the current market situation.

According to the constructed model, the forecast of subsequent values is made, taking into account the statistical characteristics of the data and the best within the model. It should be borne in mind that the remaining “overboard” factors of influence can be very significant. An adequate model is a good forecast. You should focus on a qualitative, rather than quantitative, forecast of price fluctuations and use it as a hedging signal.

The length of the price series fragment (N) is selected based on the statistical uniformity of the data (200-600 points).

Noise filter parameter Noise HF Limit: the longer the row, the smaller the filter parameter. For short fragments, the estimation of statistical properties is performed with an increased error-the level should be increased.

In the configured model, the number of prediction points with an acceptable prediction probability is 10-30 points. For a long-range forecast, it is reasonable to use a coarser timeframe. The simultaneous use of two indicators with different parameters (N=400 and 200) will indicate the divergence of the influence of factors associated with different time scales, or confirm the forecast.

P.S. The implementation using the FFT algorithm, which allowed to increase the length of the fragment and significantly speed up data processing, is presented in the SSA Fast Trend Forecast.