1 · Trading idea & research
Turn a vague hypothesis into a precise, testable edge before you write a line of code.
Every robust system starts as a falsifiable hypothesis — not "I think mean reversion works", but a claim specific enough that a backtest could prove it wrong. This stage is cheap and the highest-leverage: a bad idea, beautifully implemented, is still a bad idea.
From hunch to hypothesis
A good trading hypothesis names four things:
The edge (why it should work)
An economic or behavioural reason a price move should partly reverse or continue — e.g. forced selling by index funds overshoots, then reverts. If you can't say why, you're curve-fitting, not researching.
The signal (how you measure it)
The concrete, computable trigger: z-score of price vs. its 20-day mean ≤ −1.25. This becomes an indicator later.
The universe & horizon
Which assets, which timeframe. Liquidity, borrow availability and data quality all depend on this. A 400-name liquid US-equity universe behaves nothing like 5 micro-caps.
The exit & risk
How the trade ends: reversion target, stop, time-stop. An edge without an exit rule is half a strategy.
Write it down first
Before coding, write the hypothesis as one paragraph. When the backtest later disagrees with you, that paragraph tells you whether you found a bug or killed the idea — the two look identical in the equity curve.
Ground it in literature
State-of-the-art systems rarely spring from nothing. Anchor your edge in published evidence so you inherit decades of scrutiny — and know the documented failure modes. Malgot's seeded Statistical Mean Reversion strategy, for example, cites Avellaneda & Lee (2010) and Poterba & Summers (1988). Standing on a known result means the question shifts from "does this exist?" to "does it still exist, net of costs, in my universe?" — a far more answerable question.
Avoid the traps early
- Look-ahead bias — using information a bar could not have known yet. Decide now that signals fire on data up to and including the current bar, and act on the next one.
- Survivorship bias — testing only on today's tickers ignores the ones that delisted. It flatters every long-biased result.
- Multiple-testing — if you try 200 variations, some will look great by luck. Decide your hypothesis before you see the results, and reserve data for later stages (Robustness).
In Malgot today
Research itself happens in your head and your notes — Malgot's job starts once you have a testable claim. Two things to prepare here:
- Asset universe → create an Asset Group for the tickers you'll trade.
- Data availability → the data engine ingests OHLCV; make sure your universe and date range are covered before you rely on a backtest.
Next
Once your hypothesis is written and your universe is defined, you're ready to express it as code.