From Discretionary to Systematic: How to Codify Your Trading

A practical path for discretionary traders who want to turn conviction into testable, repeatable strategies, built around a four-step research loop.

If you have traded discretionarily for a few years, you already own the hardest asset in this business: real intuitions about how the market behaves, earned by watching it with money at stake. You have noticed that certain names bleed after certain filings, that options seem rich going into certain events, that some setups keep rhyming. The problem is not your ideas. The problem is the format they are stored in.

An intuition that lives in your head cannot be tested, cannot be sized consistently, and cannot be run while you sleep. Codifying your trading is the act of moving those intuitions into a format where a computer can check them against history and then execute them the same way every time. This post walks through what that translation actually looks like, using the same four-step research loop that the Systematic Trading with Market Data curriculum repeats on every strategy it teaches.

What you already have, and what it is missing

Discretionary experience gives you two things a textbook cannot: a feel for which market behaviors are real, and a memory of what it costs to be wrong. Those are exactly the raw materials systematic research runs on. Every quantitative strategy started as somebody's hunch.

What discretionary practice does not give you is a way to know whether a hunch is an edge or a memory. If you traded a setup twelve times and it worked eight, you have an anecdote, not a distribution. You do not know whether the wins came from the setup or from the regime you happened to trade it in. You do not know the tail: the one instance in fifty that gives back a year of gains. Systematizing is not about replacing your judgement with code. It is about subjecting your judgement to evidence before the market does it for you at worse prices.

The format problem: fuzzy claims cannot be tested

Say your conviction is "earnings options are usually overpriced." As a trading instinct, that might be pointing at something real. As a research claim, it is untestable: it names no population, no measurement, and no threshold for being wrong. Compare it to this version: for large-cap US names, the earnings move implied by the options market is, on average, larger than the move that gets realized. Same instinct, but now it has a population, a quantity, and a direction. A computer can check it. That rewrite is the entire trick, and everything else is plumbing.

Most discretionary convictions can survive this translation. "Stocks that announce dilutive offerings keep falling" becomes a claim about the average post-event drift across every offering filed in a window. "Dividend names snap back after the ex-date" becomes a claim about recovery frequency across the whole calendar. If a conviction cannot be rewritten this way, that is diagnostic too: it usually means the pattern has no defined population, which is a polite way of saying it may not exist.

The four-step loop that does the conversion

The procedure for testing a codified claim has four steps, in a fixed order: hypothesis, data, measurement, attack. The free lesson on the four-step research loop states them precisely; here is the short version aimed at a discretionary trader making the switch.

• Hypothesis. Write the claim down as one sentence with a yes-or-no answer and a magnitude, before touching any data. This ordering is the cheapest honesty mechanism in research: look at data first and it will happily suggest a claim it already supports.
• Data. Pull the full event or series history, point-in-time. Every record must reflect what was knowable on the day, not what was revised or backfilled later. A dataset polished with hindsight makes any strategy look brilliant.
• Measurement. Compute the few numbers that answer the claim: the event count, the average effect, the hit rate. Resist the forty-chart impulse. The hypothesis was one sentence; answering it takes a handful of numbers.
• Attack. Try to kill your own result. Check the tails, split the sample by regime, question the sample size, subtract realistic costs. If the effect only lived in one volatile year, it is not an edge, it is a memory.

The attack step is where discretionary habits fight you hardest, because it inverts the reflex you have spent years building. The discretionary move after a good result is to press it. The systematic move is to turn adversarial and go looking for the boring explanation. Our checklist for reading a backtest honestly expands this step into a permanent routine.

A worked translation, start to finish

Suppose your discretionary observation is that small caps announcing last-ditch share offerings tend to keep sliding. Codified: for US-listed companies that file a dilutive offering, the average return over the following weeks is negative. Data: a point-in-time history of dilution events, so each record shows the filing as it appeared on the day. Measurement: event count, mean post-event drift, fraction of events that closed lower. Attack: are the losses concentrated in a handful of collapses, does the effect hold outside one bear year, and does borrow cost eat the short-side edge?

Notice what happened to your role. You still supplied the idea; the market knowledge was yours. What changed is that the claim now gets settled by several hundred events instead of the dozen you personally remember, and the answer comes with a magnitude attached. If the result survives, you have a rule you can run every morning. If it dies, the loop killed the idea for the cost of an afternoon instead of a drawdown.

What changes about your trading day

The day-to-day shift is from watching names to querying populations. Instead of monitoring a watchlist and waiting to feel something, you run a screen: which names filed the relevant event, which pass the filters, what does the rule say about size. The decision was made during research; the morning is execution. Traders coming from discretionary backgrounds often describe this as boring, and that is the point. The excitement was never free. You were paying for it in inconsistency.

What you should not throw away is the idea engine. Screen time, market reading, and the pattern sense you built discretionarily remain your source of hypotheses; the loop is just the refinery. The traders who make this transition well keep generating hunches and simply stop trading them raw. For a concrete picture of what a first codified strategy should look like, see what to build first and what to skip.

Where to start

The fastest way to internalize the loop is to run it, not read about it. The first module of Systematic Trading with Market Data is free with no account required, and it builds exactly this foundation: why edges exist, which events can be studied as populations, and the four-step loop itself, with code you run in the browser against real market data. The first lesson takes about ten minutes. The full curriculum runs roughly 18 hours and repeats the loop across volatility, event-driven, and flow-based strategies until it becomes the only way you can think about a trading idea. When you are ready for the data side, the Alphanume datasets are the same point-in-time event histories the course queries.