QUOTE:
to see if any stock in my watch list meets the criteria or does just do back testing to see my strategy would have worked.
We are talking about both tiers 2
and 3 here. I would use
separate strategies for both tiers 2 and 3.
Tier 2, as discussed in Post# 2, is employing a strategy which only needs to perform a trade on the "off-the-chart bar" (i.e. the Bars.Count bar, not the Bars.Count-1 bar) to generate an Alert. Recall, Alerts are only generated for the off-the-chart bar. As a result, you don't really need a trading FOR loop to perform that. All that's important is to perform a Buy on the off-the-chart bar; no other trades are required. The purpose of this tier 2 "strategy" is only for
screening, not trading. In your case, you're using the RSI indicator to screen with. In practice, you would probably logically "AND" a combination of criteria.
In Tier 3, you are employing a "real" WL trading strategy with a real trading FOR loop to trade throughout the Data Range. The real goal here is to use the WL Optimizer to set the PV model parameters for best results. After setting those parameters for that strategy, you need to look at the trading equity curve to be sure it's
positive going. If it is, then--and only then--do you add that stock to your trading dataset for that specific trading strategy. Obviously, if the equity curve for that stock/strategy pair is negative going, then it's a bad match up.
Some trading strategies work well for some stocks and not others, and vice versa. That's what you have to figure out for tier 3 when using the PV Optimizer.
And for tier 3, you also want to make sure you have enough DF for error in your parameter fit. Obviously, if you only have 3 trades in a 3 parameter model (i.e. zero DF for error), then those modeling parameters could have just been fitted by happenstance. You really need to have an "over determined" system (from an numerical analysis point of view) in a randomly varying process so that there are around 5 DF for "random" error (i.e. 5 "extra" trades--or more). Trust what your stat course taught you about modeling randomly varying processes.
Some WL users are paranoid about having enough DF for error in their models, so they pick Data Ranges of ten years or more. That's probably over done and undesirable since the market changed significantly in 2015 when the Fed started raising interest rates. I would set my Data Range back to 2015, but you may find some stocks won't trade enough to give you 5 DF for error going only that far back. If that's the case, I would just avoid those stocks ... for now.