Discretionary versus systematic

Some traders follow rules a machine could execute. Others make judgements bar by bar. Most sit somewhere in between, using rules for some parts of the process and judgement for others. Backtesting works cleanly on the first group, awkwardly on the third, and not at all on the second. Knowing where your process sits tells you what kind of testing is possible and what kind of evidence you can honestly claim.

The spectrum

A systematic trader can backtest ten years in an afternoon. A fully discretionary trader cannot backtest at all in the strict sense, because the "strategy" is the trader's brain, and the brain in 2019 did not have the information the brain has now. What a discretionary trader can do is replay: step forward one bar at a time, decide, record, and score. That is slower and less certain, but it is honest, and Module 7 covers the mechanics.

What discretion costs you in evidence

Every element of discretion reduces what a backtest can claim.

• If entries are discretionary, the backtest tells you nothing about entries. You can still test the exit and sizing rules by applying them to your actual historical entries.
• If exits are discretionary, you cannot compute expectancy from the rules; you can only measure it from your log.
• If a "veto" is applied to systematic signals, you need a log of every vetoed trade and how it would have done. Most traders who keep this log discover that their vetoes lose money on average, because the skipped trades were the uncomfortable ones and the uncomfortable ones are often the best.

The veto log

If you want to keep a discretionary veto on a systematic strategy, run the experiment properly. For every signal the system generates, record whether you took it, why not if you did not, and the outcome in R either way.

After fifty signals you have two expectancies: the system's and yours. If yours is higher, the veto is adding something and you should try to turn the reason into a rule. If the system's is higher, the veto is costing you and the honest fix is to remove it.

Key idea: Discretion is not wrong, but it is untestable in advance. The more of your process is written as rules, the more of it can be checked on history and the less of it depends on you being right in the moment.

Converting judgement into rules

The path from discretionary to systematic is usually done one component at a time, starting with the parts where judgement adds the least.

1. Sizing first. Almost no one's sizing judgement beats a fixed-fraction formula. Make it a rule.
2. Stops second. Where you place the initial stop is usually describable: below the signal bar low, 2 ATR, below a swing. Pick the description that matches what you already do.
3. Filters third. "I only trade this when the market is trending" becomes an ADX threshold or a moving-average slope condition. It will be a worse filter than your eye on some days and a better one on others; test it.
4. Entries last. This is where discretionary traders usually believe their edge lives, and sometimes they are right. Keep the setup as a rule and the trigger as judgement while you gather the veto log.

Each conversion gives you something testable. Even a strategy that remains 40% discretionary gains a great deal from having the other 60% verified.

Where discretion is legitimately better

There are things rules do poorly: reading a news event in real time, recognising that a data feed has gone wrong, noticing that a stock is halted or that a contract is about to expire. A systematic trader still needs a discretionary override for operational problems. The distinction is between overriding the system because something is broken, which is necessary, and overriding it because the signal feels wrong, which the veto log will usually show to be expensive.

Try it: List every decision in one of your trades from idea to exit. Mark each as R (rule), J (judgement), or M (mixed). Count them. The R decisions are what your next backtest can cover; the J decisions are what only a veto log or a replay can cover.

Recap

• Strategies run from fully systematic to fully discretionary; testability falls as discretion rises.
• Discretionary components cannot be backtested, only replayed or logged.
• Keep a veto log for every skipped signal; compare your expectancy with the system's.
• Convert judgement into rules in order: sizing, stops, filters, then entries.
• Keep discretionary overrides for operational failures, not for signals that feel wrong.