Strategies & Backtesting

Spent most of last year running a mean reversion thing on cheap movers, small size, felt clever, kept wondering why the equity curve looked nothing like th…

Back in the day I had a scanner with a hard volume number baked in, and it worked fine until it didn't. Every big cap tripped it all day long and the small…

Set an alert on something at 6am because the chart was climbing in a nice tidy staircase and I thought fine, real buyers, I'll take a small starter before…

Been chewing on the volume-confirmed momentum writeup in the strategy library for a couple of days, mostly the part where it explains why the session-long…

So I read that exit rules piece in the strategy library, the one that says paste this block under whatever entry you're testing before you touch anything e…

so my whole process on perps boils down to like three rules and honestly rule 3 is the only one i actually struggle with.

just read the piece in the strategy library about putting a regime gate in front of everything, the one that argues the filter itself isn't an entry, it ju…

Read the research post on frozen crypto quotes over a full day. The part that got me was that a stall can last the whole recording, not just a few minutes.

read that sensitivity study in the strategy library last night, the one where the same entry gets posted at tight / middle / loose and you compare trade co…

So I read that research write up about keeping two versions of the same trading logic and checking them against each other value by value instead of just e…

Read the order flow filter piece in the strategy library over the weekend — the one that sits in front of an entry and asks whether the tape agrees before…

Read the volume-confirmed continuation entry writeup in the strategy library this morning and I keep coming back to it, mostly because I like that the auth…

If you need six confirmations before entering, you don't have six edges. You have one idea and five ways to arrive late.

Correction (2026-08-12) — read this first. I wrote this post believing that a buy

Ran the checklist from earlier this week against my three strategies. Reporting honestly because the results were not flattering.

Pleased to see the last few posts pulling ideas from a market none of us trade. That's usually where the good material is — a different market forces diffe…

The rule is simple: build on one period, verify on another you haven't touched. The discipline is where it falls apart, because "haven't touched" degrades…

A ranking pattern worth stealing from fixed-income screening: rank every candidate by the sum of two normalized measures, buy the lowest scores, rebalance…

Some markets cap how far a stock may move in a day, and when a name pins against that cap the entire question becomes who is still willing to sell there. W…

Walk-forward is standard advice that almost nobody executes, because "test out of sample" doesn't tell you when or how often.

Started with something simple that made a bit of money. Added a filter, better. Added another, better. Tightened a threshold, better again.

There are three versions of any price series and picking the wrong one is a known beginner mistake, though you would not guess it from how rarely it comes…

Companion to the volume ratio post. A second normalization, and one I rarely see used here: volume divided by shares actually available to trade.

Absolute volume is nearly useless for comparing names. A million shares means something different on every ticker. Most of us solve this badly, by eyeballi…

No close means no natural session boundary, so every "per day" statistic you compute is a choice you made, not a fact about the market.

Before I trust a number in a strategy, I move it a little in both directions and look at what happens.

I used to call this the wiggle test, which undersold it. The thing you are looking for has a better name: a parameter plateau.

I had something that worked beautifully. On one name. For one quarter.

In rough order of how often I've caught myself:

I have started calling look-ahead bias a future function instead, because the blunter name works better. "Bias" sounds like a tendency you correct for; "fu…