This process took off after 9/11. Today, in fact every day of the year, the NSA, CIA and related agencies intercept enough email, cell phone voice traffic and whatever else they pick up to fill the Library of Congress 5 times over. To do much with this manually in raw form would require several million human readers and listeners...totally impractical. So they do an immensely more advanced form of automated WF architecture identification before most of it gets the attention of a human intelligence analyst. I'd love to get my hands on the software and computer resources they have for this as by comparison the markets are much less complex. Instead of looking for the voice prints of, what 30,000 possible terrorists around the world you have only three major kinds of markets: stocks, futures and Forex. Aside from stocks each of these has only a few dozen significant price time series. Plus you don't need to look at every market, but only those you choose to trade. One problem is complexity and computer resources required for the predictive analytics. It takes 4 high-end Dell workstations a week to remodel the complete process and this has to be done every couple of months, depending on bar length used, to stay in sync with market evolution.
The crossover? How do you explain all the times the crossover occurred yet you didn't take the trade? Also, the best case scenario for entering on a crossover is entering at the first tick of the first bar AFTER the crossover, because due to the way a candle forms, the crossover is not guaranteed until after the bar has printed... it can change back and forth many times while the bar is still forming. I must have missed something from your chart. The entries you marked were good, but there were also entries that follow those same rules that you ignored. Why?
Looks awesome in that pic. How does it do in real time? You know if it has high accuracy you could average down into those trades when price doesn't reverse quite right away and double your money. IB people flame me for suggesting average down.
The issue is not indicators but the following: Trend Acceleration Volume +/- Momentum ATR Velocity Overbought and Oversold Potential for return to equilibrium So I use: Candles A special coded ATR TSI Chandes momentum Oscillator CCI Volume And an addl leading indicator using a variation of the ATR with averaged EMA smoothing in currencies futures. And an addl up down Volume indicator. There isn't any indicator that I feel predicts acceleration well that isn't purely a subjective and intuitive guess however and that are mostly derived from multiple indicators that predict velocity.
I'll probably know how or if it works in about a month. The computational complexity makes turning it on live a major task so I do extensive backtesting before taking that final step.
Nice patience and I mean that wholeheartedly!! It has taken me years to program what I do and it was worth the wait.
I always do until I have a series of statistical metrics for demo performance in relation to out of sample performance and then the same for live trading to OOS performance. More then one promising system architecture has failed in live trading due to fills not being would back testing would assume to be possible or becasue the market evolves into perviously unknown pattern sturctures that the design can't decode sucessfully