Security price time series clearly are causal. Prices are sufficiently correlated over all time frames to encourage the trader to act as if they are not random. Over day trading time frames both their signals and their noise are sufficiently statistically stationary to apply classical tools. The associated volume time series suggest a weakly non-linear first order differential equation relationship with price which is exploitable. The latencies and inaccuracies in internet reporting of price and volume are well within tolerable limits on day trading time intervals. The only serious issue is that third and higher order derivatives of price with time are significant.
Therefore virtually any modern estimation theory approach should provide a viable approach to trading. For example, statistical estimation, Laplace frequency domain filtering, and Kalman filtering should work. But such approaches suffer from computational issues on PCs for real-time trading. Therefore my solution is to fall back on classical analog servomechanism theory. This is an appropriate tool for quasi-stationary statistics and for uncertain dynamic causalities. And it is an appropriate mental modality to work in because the objective of trading is to have the trader follow the market mechanically, objectively, and without passion.
So I treat price as the raw signal and design feedback outputs to drive the trader to the desired state. Attached is an example system designed to identify and ride potential all-day holds in NQ on a five-minute time frame. Green in the helper pane is a long signal. Red is a short signal. Orange means to stand aside. The chart shows an example full day in U.S. West Coast time. Of course this was chosen because it worked so well, but the result is not unusual. The method is scalable to longer time frames, but is highly problematic for charting faster than 30 seconds. Later I will show intraday trades and the system I use for investing.
Therefore virtually any modern estimation theory approach should provide a viable approach to trading. For example, statistical estimation, Laplace frequency domain filtering, and Kalman filtering should work. But such approaches suffer from computational issues on PCs for real-time trading. Therefore my solution is to fall back on classical analog servomechanism theory. This is an appropriate tool for quasi-stationary statistics and for uncertain dynamic causalities. And it is an appropriate mental modality to work in because the objective of trading is to have the trader follow the market mechanically, objectively, and without passion.
So I treat price as the raw signal and design feedback outputs to drive the trader to the desired state. Attached is an example system designed to identify and ride potential all-day holds in NQ on a five-minute time frame. Green in the helper pane is a long signal. Red is a short signal. Orange means to stand aside. The chart shows an example full day in U.S. West Coast time. Of course this was chosen because it worked so well, but the result is not unusual. The method is scalable to longer time frames, but is highly problematic for charting faster than 30 seconds. Later I will show intraday trades and the system I use for investing.
