All,
I have an example of some classification data that would be theoretically possible to get with a P123 download (hypothetical as just an example). Maybe you would want to use IF x>0 then 1 else 0 in Excel to convert continuous returns into a class. I am not against using Excel for some of this. A lot of the articles on trading with technical data focus on whether to go long or short on an equity and use classifications (if it is necessary to explain a ‘case use’ to not get excluded from the forum).
But with the question of whether to use this strategy I might ask ChatGPT this:
The answer:
Actual Positive Actual Negative
Predicted Positive 1500 1021
Predicted Negative 1000 1600
Here are the updated metrics for the stock market signal along with the result of the chi-squared test for statistical significance:
Accuracy: 60.54%
Recall (Sensitivity): 60.00%
Specificity: 61.05%
F1 Score: 59.75%
P-Value (Chi-squared test): 4.67×10−514.67×10−51
Interpretation of Statistical Significance
The p-value obtained from the chi-squared test is extremely close to zero, indicating that the differences observed in the confusion matrix are highly statistically significant. This result suggests that the signal’s predictions are not due to random chance but are likely influenced by underlying factors or patterns in the data.
In other words, the signal seems to provide meaningful information about the market’s direction, and its performance is not merely a result of random fluctuations.
If you have any further questions or need additional insights, please let me know!
Back to jrinne: My apologies, but I genuinely do not understand how that could possibly be a bad thing for P123.
I know feature request are usually ignored at P123 (so I will not make one). But if I worked at P123, I would probably decide (on my own) where to include a few confusion matrix downloads and let ChatGPT run the statistics, Just because it is not that hard and I wanted to promote machine learning on the P123 platform.
Again though with my main point: statistical calculations are no longer even desired in many instances (e.g., not everyone cares about F1 above). Just let ChatGPT do it without having to commit-to or exclude a particular method or take a side in a debate on the forum!!!
And ChatGPT could at least start to answer any questions. As far as advice, it usually recommends caution with a strategy and suggests further studies, tests etc. I think it does well in that regard even if the advice always boils down to using cross-validation. Maybe ChatGPT could help an interested newbie with that. And I am not sure that there are not some designers that could use that advice. I am pretty sure they would not have to take it.
Jim