Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
You can start with the persistence model and use the previous day midpoint or perhaps the close price as your benchmark. That would be the benchmark method your formula needs to beat and if you can't beat it suggest the market is a random walk and/or your methods have no additional predictive value. You can use a metric such sum of squared distance between your actual value and your predicted value to see which method is best.
T A S software can project a dynamic profile for the next time frame if things fall in to the algos eye sight. value high low and poc. you might be able to take the rth open and over lay yesterdays profile. just set them side by side drawing a level line through the opens... it is nuts i know but you have to start some where..
I was suggesting 2 different methods, either using the midpoint of the prior day or the close of the prior day. Theoretically, the last price should have the most information, i.e. so that's going to be the best bet of your midpoint. However, I don't know if some information from the volatility might be missing from that that the midpoint from the prior day might capture. It is a great question though! Right, you can test it by simply shifting your data and taking the difference from the actual midpoint and summing the value. The formula I gave was the mean squared error but you could use the absolute difference or another error measure if you desired.
I coded up a quick verification in Easylanguage and using the close of the prior day is significantly better than using the midpoint for predicting the midpoint, assuming I didn't make any mistakes (big assumption). The code is below and I've linked to an instructive video if you want to learn more.
The average error in points is around 11 points for using the close over the past couple years I tested. This would mean on average using the close, the midpoint will occur within +- appx 6 points. That's your accuracy level. Using the prior midpoint, I get some ridiculous number which means I probably made a mistake. I updated to calculate using the simple MAD (mean absolute deviation) and get 8 points for the close and 14 for the prior midpoint. That would put you on average +-4 off using the prior close. PS: If someone wants to explain the differences in mean squared error vs absolute average error -- it would be illustrative for me!
Update: Found the mistake. You must wait until CurrentBar=2 because there isn't a previous value for the first bar. It did not change the findings though.
remember i said this is nuts so do not lola me to bad...you have yesterdays RTH profile on the ES ...part 1... this morning the ES RTH opens and the chart is blank just a * for the open ,,,that is part two. you move yesterdays profile in place where today RTH OPEN an yesterdays RTH close are on the same price level ...part 3 ...now yesterdays profile is your projection profile for to day... look for the poc ...that is the project poc for to day. like i said it sounds nuts but you have to start some where...what you are saying is day 1 will look like day 2 just at different levels.
Tested the floor pivots (h+l+c)/3 and using the prior close still had a lower error. I'd expect the open to have an even lower error but can't test it easily. You could try to use machine learning to do better but the tradeoff is the introduction of model complexity. But, there is hope, I have came up with a model that beats the simple model nearly every single day for the past 10 years or so but only by a tiny amount.