@ddavis534 on GBPUSD | PriceONN Community
Hey all, been tweaking my algo for GBPUSD lately and I'm curious if anyone's had experience using Kalman filters for noise reduction in their models? I've been reading about it, but the implementation seems a bit tricky. What parameters did you find most impactful, and did it improve your backtesting results significantly? I'm particularly interested in its impact on reducing whipsaws during range-bound sessions like we've seen recently.
Replies
So, about those Kalman filters... I'm thinking specifically about applying it to the RSI and MACD indicators to smooth out the signals. The raw signals often generate false positives, especially with GBPUSD's tendency for sudden reversals. I've seen some papers suggesting that a combination of Kalman filtering and adaptive moving averages can improve signal accuracy. But I'm also wondering about the computational cost – could slow down real-time execution too much. Anyone got any Python snippets they'd be willing to share or point me to?
B
@ddavis534 Kalman filters are interesting, but be careful of curve fitting. Have you tried using a simple moving average to smooth the RSI first? Might be easier to implement and test.
@ddavis534 Interesting idea about the Kalman filters. I'm still learning, so I've mostly stuck to simpler moving averages for smoothing. What kind of time frame are you using with your RSI when applying the filter?
O
Hi @ddavis534, interesting approach with the Kalman filters. I've explored them in the past for signal smoothing, especially on indicators like the Stochastic RSI where whipsaws can be particularly problematic during choppy market conditions, which we're seeing a lot of with GBPUSD lately. My primary concern with Kalman filters, and perhaps you've encountered this, is the sensitivity to the initial state and the noise covariance matrix (Q and R). Getting those parameters optimized correctly for a specific asset and timeframe can be an extensive process, often requiring significant historical data for calibration. Have you found that the computational overhead of a Kalman filter outweighs the benefits compared to simpler smoothing techniques like exponential moving averages, especially when dealing with high-frequency data or when speed is critical for your execution?