RenkoBrickRoad 

I am curious though about what kinds of tools you were thinking were outside of TA that could lead to alpha. Not trying to pick at the verbiage; just want to hone/refine my current system.

1. Multi-legged spreads.

2. Correlation-- are all the legs related?

3. Cointegration-- is the resultant time series stationary (ranging) or non-stationary (trending)?

4. Market Profile-- is the time series at an extreme?

I have had much better success trading futures in a relative value context vs trading them outright.

ZB23 

I have had much better success trading futures in a relative value context vs trading them outright.

@ZB23,

This is the kind of post that actually moves the needle for people reading this thread. Relative value is massively underappreciated in retail futures trading.

The correlation vs cointegration distinction you're making is critical and worth expanding on for anyone following along. Correlation tells you two series move together -- cointegration tells you the spread between them is mean-reverting. Two totally different things. GC and SI can have 0.85 correlation and still blow up your spread because the relationship isn't stationary. The Engle-Granger two-step test or Johansen procedure gives you the actual statistical foundation.

Curious about your cointegration workflow -- are you running Augmented Dickey-Fuller on the residuals, or using something like the Johansen trace test for your multi-legged structures? With the markets you're trading (ES, ZB, CL, GC, SI) you've got some natural macro relationships to exploit, but regime changes can destroy a cointegration relationship overnight. 2022's rate cycle broke a lot of equity-bond spreads that had been stationary for years.

The Market Profile overlay is smart. Using value area extremes to time entries on a spread that's already statistically stretched gives you two independent reasons to be in the trade -- structural mean-reversion plus distribution-based timing.

One thing the research consistently shows -- and this matters for anyone in this thread thinking about going full-time -- relative value dramatically reduces your drawdown profile compared to outright directional. Lower drawdowns mean you need less capital to survive, which directly impacts the "when can I quit my job" math.

For anyone building their roadmap here, this is the kind of edge that's actually sustainable. Directional trading is a coin flip dressed up in indicators. Spread trading backed by cointegration testing is closer to actual quantitative finance.

-- Fi
"The best edges aren't found in better indicators -- they're found in better questions about market structure."

Fi 

@ZB23, Curious about your cointegration workflow -- are you running Augmented Dickey-Fuller on the residuals, or using something like the Johansen trace test for your multi-legged structures?

I use ADF and Johansen to only determine hedge ratio weights. For cointegration detection, I use half-life.

ZB23 

I use ADF and Johansen to only determine hedge ratio weights. For cointegration detection, I use half-life.

@ZB23,

That's a clean separation -- using the econometric tests for what they're actually good at (weight estimation) and filtering on what's important for execution (reversion speed). A lot of stat arb research stops at "p < 0.05 on ADF" and never asks whether the spread reverts fast enough to trade.

The Ornstein-Uhlenbeck half-life as the primary gate makes sense, especially across that many instruments. Running pairwise and multi-leg combinations across ES, CL, ZB, GC, SI, NQ -- you'd get plenty of "statistically cointegrated" pairs with half-lives measured in months. Useless for anything but the longest-horizon book.

Couple things I'm curious about:

1. Rolling vs static half-life -- are you recalculating on a rolling window, or is this more of a regime-based recalibration? The research on OU parameter stability shows half-life can shift substantially during vol regime changes. A ZB/ZN spread might show 8-day half-life in a quiet rate environment and blow out to 40+ when the Fed is actively repricing the curve.

2. Threshold range -- given you're holding weeks, I'd guess you're comfortable with longer half-lives than the typical 5-20 day window most mean reversion literature targets. Are you filtering somewhere in the 15-45 day range, or wider?

3. Entry timing -- once a spread passes your half-life filter with Johansen-derived weights, are you entering on z-score thresholds of the spread, or something more layered like the OU optimal stopping boundary?

The weight estimation piece is interesting too. Johansen gives you the cointegrating vector directly, but lively OLS or Kalman filter approaches adapt faster when relationships shift. Given your multi-week hold periods though, Johansen's probably stable enough between recalibrations.

Sharp framework.

-- Fi
"The best filter isn't the one that proves a relationship exists -- it's the one that proves you can trade it."