How HunterML works
HunterML is my machine learning project for futures trading on NinjaTrader 8, and it is in training. It has not shipped, and I have not published a single performance figure for it, because there is nothing yet that would be honest to publish. This page is the plain-language version: the idea, the standard it has to clear, and where it actually stands. No dashboards, no results, no launch countdown.
A machine learning project, still in training
The twelve strategies I sell today are rule-based. Every entry, every filter, and every exit is logic I wrote by hand, and you can read all of it. HunterML is a separate experiment pointed in the other direction: instead of me writing the rules, a model learns patterns from futures market data.
It is one project, in training, not a product. It does not replace the twelve live strategies, which stay exactly as they are. It is not for sale, and there is no date attached to it. When people ask what it does, the honest answer today is that it is learning, and I am still finding out whether what it learns is worth anything.
The idea: reading market regimes
Markets do not behave the same way all the time. Some sessions trend cleanly in one direction. Some chop sideways in a tight range. Some are quiet and some are violent. Traders have names for these conditions, and the general term is a market regime.
A rule-based strategy has a fixed idea of the conditions it wants. A breakout strategy wants expansion and momentum, and it has a harder time when the market is ranging. That is exactly why the live strategies lean on session and volatility filters to stand down when conditions are wrong for them.
The idea behind HunterML is narrower, and I think more honest, than the pitch of an AI that just trades. It is a model that learns to classify the current market regime from futures market data, so a strategy can be told when conditions suit it and when it should stand aside. The machine learning is not there to conjure a magic entry. It is there to answer one question well: what kind of market is this right now?
Whether classifying regimes that way actually helps is the entire question, and it is not answered yet. Believing an idea is elegant is not the same as proving it holds up. That gap is what validation is for.
How I will know whether it works
I hold HunterML to the same standard as everything else on this site. Nothing gets published, and nothing ships, until it survives these steps in order.
- 1
Backtest
Does the idea hold up on historical futures data at all? This is the easiest bar to clear, and the easiest one to fool yourself with, because you can keep tuning a model until it looks flawless on data it has already seen.
- 2
Out-of-sample validation
The real test: does it hold up on data it was never trained on? A model that only looks good on its own training data has memorized that data and learned nothing general. This is the step that catches curve-fitting, and it is the step HunterML has not cleared yet.
- 3
Live simulation
Running forward in real market conditions on a simulated account, with no money at risk, to see whether the earlier results survive contact with live data and real execution.
I will not publish results, and I will not ship, until the out-of-sample step holds. Machine learning makes it easy to build something that looks brilliant on paper and behaves terribly with real money. The whole reason to move slowly here is that a model graded only on its own homework is exactly the kind of thing that produces a beautiful backtest and a painful live account.
What HunterML will not do
The most useful thing I can tell you about a product that is not finished is what it will never be, no matter how the training goes.
Send you signals.
Like the twelve live strategies, HunterML would run locally on your own NinjaTrader 8, against your own broker connection. There is no alert feed to follow and no master account. If it ships, you run the software; I do not run a service.
Be a black box you cannot inspect.
When it ships, the methodology ships with it. You will be able to read how it classifies conditions and when it stands a strategy down, the same way you can read the rules on the strategies today. I am not going to hand you a sealed box and tell you to trust the AI.
Guarantee anything.
Machine learning does not remove risk from futures trading. It cannot promise an outcome, and I will not imply that it can. Futures trading carries substantial risk of loss no matter what is making the decision.
Trade on autopilot for free money.
There is no version of this, shipped or not, that turns a market into an ATM. Any product that promises that is selling a fantasy. HunterML is software I am still building and testing, and that is all it is.
Where HunterML stands
It is in training. I have not published any results, because the out-of-sample validation that would make those results meaningful is not complete. I do not announce launch dates. It ships when it earns it, not when a calendar says so.
When it does ship, it ships the way everything here ships: documented methodology first, honest results carrying the same hypothetical-performance disclosures as the live strategies, and only then a product you can buy. If you want to know the moment that happens, the launch list sends one email and nothing else.
Futures trading involves substantial risk of loss and is not suitable for every investor. No performance claims are made for HunterML. Read the full disclosures.