So I was able to code the NL scoring and I ran it for a variety of OR's and trading hours and than ran ran some correlation analysis on all those scores vs a variety of % changes in the underlying IE change tomorrow, 3 day, 7 day change etc... Some interesting takeaways... 1. The numberline score with the most predictive value value tends to be the shortest numberline. 2. Numberline score and % change for the day that generated that score are very weakly correlated. In fact for gold the most predictive numberline time period is the least correlated with that days close to close % change. This is a very good thing because it shows that NL is looking at the markets in a unique way vs measuring pure momentum.
Here is a scattergraph of short NL on gold vs 3 day pct change. Of the greatest interest I think is the noticeable lack of big down moves with a NL > 5, and a less noticeable lack of big up moves with a NL <-4.
Way back in this thread Shan reported his findings on the 5D NL rate of change. A couple of years ago I spent a long time investigating that, manually, since I can't code and at my age I don't want to spend the time to learn (it's all opportunity cost for my time nowadays). @Robert Yanks can point you to the page, he actually refreshed my memory on that. Since I know options, I used the same concepts and developed 2nd-4th order derivatives of the 5D NL and they are pure gold in flagging moves. Once the move is underway, less useful with the ebb and flow.
Here is a list of fed minutes scraped from the fed site *Not checked for errors. For educational purposes the attached is the code in R that I used to scrape it. I threw it together quickly as I don't have much time but you could probably code something similar in Python. Hope it helps.
So I was playing with some filters on gold NLs and found some very promising ones based on that scatter graph I posted. Took same filters and applied them to ES without any tweaks. First time I've messed with ES at all really, and got these results. I'm going to look through and make sure I didn't accidentally let it see into the future or something but needless to say this is encouraging. I know the source of the 0 spikes btw. Haven't bothered cleaning it up but it won't change the distro once I get rid of it.
They are all of the same subset of days. X axis is % change over the designated time period following the NL I filtered for, Y axis is count. Having said that, I'm not naïve so right now I'm trying to find my mistake.
It's like the markets. It ebbs and flows, but it is never wrong and is always riveting when it moves.