What quantitative trading is: rules, evidence and realistic expectations

Updated

What you will learn

  • Explain the difference between discretionary and rule-based trading.
  • Name what written, tested rules can and cannot do.
  • Set realistic expectations for research results and for the time it takes.

Quantitative trading has an image problem. The name suggests physics doctorates, server rooms next to the exchange and returns that no one can explain. At its core, though, the idea is modest: decide in advance, in writing, exactly what you will buy, when you will sell and how much you will risk, then check on past data whether those rules would have done anything useful before you trust them with money. This course is about that modest version, the one an individual trader with daily price data can actually practise.

Rules instead of judgement

A discretionary trader looks at each situation and decides. The decision may draw on years of experience, but it is made fresh every time, and two similar charts can lead to different trades depending on mood, recent results or the news of the day. A rule-based (systematic, or quantitative) trader writes the decision down once and then applies it the same way every time.

A rule, in this sense, is a precise statement that uses only information available at the moment of the decision. “Buy strong stocks on weakness” is an intention. “Buy at the next open when the close is above the 200-day simple moving average and the 2-day RSI closes below 10” is a rule: two people applying it to the same data will always reach the same answer.

The distinction is not that one approach uses numbers and the other does not. Plenty of discretionary traders use indicators, and plenty of systematic rules are simple. The difference is when the thinking happens. The systematic trader does the thinking before the trade, writes it down, tests it, and then mostly executes. That ordering is what makes the process measurable.

The research loop of a rule-based traderSix boxes in a row: idea, rules, backtest on a training period, one look at a holdout period, small live trading with a journal, and review. A dashed arrow returns from review to idea.1. Ideawhy it might work2. Rulesentry, exit, size3. Backtesttraining period4. Holdoutone honest look5. Small livejournal trades6. Reviewkeep or dropMost ideas stop at step 3 or 4 with “No edge”, and that is a useful answerA rule-based research loop
The research loop. Each idea goes through the same steps, and most stop early: a rule that shows no edge in testing never reaches real money.

Every lesson in this course covers one step of the loop: turning an idea into rules (lesson 2), running a fair backtest (lesson 3), protecting yourself against overfitting with a holdout period (lessons 4 and 5), avoiding data traps (lesson 6), moving from a test to live trading (lesson 7), and finally running the whole process in Clear Tape (lesson 8).

What written rules can do

  • Make a method testable. Only a rule can be replayed on twenty years of prices. A discretionary method cannot be tested honestly, because its user already knows how the charts turned out.
  • Make results attributable. When a rule-based trade loses, you can tell whether the rule failed or you failed to follow it. With discretion the two are tangled together.
  • Remove some of the worst habits. Moving a stop further away, skipping the signal after three losses, doubling up after a win: a written rule leaves less room for each of these.
  • Fix risk before the trade. Position size follows from a formula (lesson 2), so the loss on any one trade is known in advance, unless a gap jumps the stop.
  • Scale your attention. A rule can scan five hundred stocks after every close in seconds; a person cannot read five hundred charts every evening.

What rules cannot do

Rules do not know the future. They describe a response to conditions, and a test shows how that response would have worked in the conditions that happened to occur. Several things lie outside any rule’s control:

What rules control and what they do notTwo columns. Rules decide entries and exits, risk per trade, which stocks qualify and the response to every signal. Rules cannot decide next year’s market return, gaps through a stop, whether a past regime repeats, or whether others find the same edge.Rules decideRules cannot decideWhen to enter and exitHow much to risk per tradeWhich stocks qualifyWhat happens on every signalWhat the market returns next yearGaps through a stopWhether the past regime repeatsWhether others find the same edge
Rules govern your own actions. The market, gaps and the behaviour of other traders stay outside them.
  • Future returns. A rule that beat buy and hold from 2000 to 2014 has shown that it did, not that it will. Lesson 4 explains how a holdout period gives a fairer, but still limited, check.
  • Losing streaks. Even a sound rule loses often. A trend-following rule may be wrong on most trades and still be worth running; a mean-reversion rule may win often and then give back months of gains in one week.
  • Edges that fade. When a pattern becomes widely known, more traders act on it and the reward tends to shrink. R. David McLean and Jeffrey Pontiff found that the returns of published stock-market anomalies were noticeably lower after publication than in the original studies. Expect the same of any rule you find in a book or online, including the classic rules in this course.
  • Your own discipline. A rule only helps if it is followed. Lesson 7 covers how to check that.

Where an individual can compete

Professional quantitative firms spend heavily on speed, data and execution. An individual cannot compete with them at the very short end, where profits depend on being microseconds faster or paying fractions of a cent less per share. The individual’s realistic territory is the other end: rules that run on daily bars after the close and hold positions for days to months.

Where an individual can competeA holding-period axis from microseconds to months. The short end is shaded as a speed race needing co-location and tick data; days to months is shaded as within reach of an individual using end-of-day data.microsecondsminuteshoursdaysweeksmonthsSpeed raceco-location, tick dataWithin reach of an individualend-of-day data, rules run after the closeHolding periodShorter horizons need faster data, cheaper execution and more capital to compete
The shorter the holding period, the more the result depends on speed and execution cost. Daily rules held for days to months are where the individual’s disadvantages matter least.

That territory also has advantages of its own. A small account can buy or sell its whole position at the open without moving the price. It does not have to report to clients every month, so it can tolerate a long flat period. And it can choose to trade only liquid US stocks and ETFs, where costs are low and data is clean. Daily data, simple rules and patience are a coherent combination, not a compromise.

Realistic expectations

Three expectations save a great deal of disappointment:

  1. Most ideas fail. Tested honestly, against a fair benchmark, after costs and on data the rule did not see, most trading ideas show no advantage over simply holding the same stocks. In Clear Tape that result is reported as No edge, and it is a useful answer: it has just saved you from trading something that does not work.
  2. A good rule is usually a modest one. A rule that matches the market’s return with smaller drawdowns, or adds a little return for the same risk, is a genuine success. A backtest that shows several times the market’s return with tiny drawdowns is far more likely to contain a mistake (lessons 4 to 6) than a discovery.
  3. Research takes time. Writing a rule takes an evening. Testing it fairly, checking neighbouring settings and then watching it trade small for a few months takes much longer. That slowness is part of the method, not a flaw in it.

None of this is a reason not to begin. It is a reason to judge your work by the quality of the process rather than by the size of a backtested number. A process that rejects nine ideas out of ten for good reasons is working.

How to apply it

Entry

Start a research project only when the idea can be stated as a rule that uses information available at the time and comes with a plausible reason why it might work (a behavioural pattern, a risk premium, a structural flow). An idea without a reason is likely to be a pattern in noise.

Stop

Decide before testing what result would make you drop the idea: for example, no advantage over the benchmark in the holdout period. Write it down next to the rule.

Target

Aim for a rule that is simple, explainable and modestly better than the benchmark on return or on risk. Treat anything spectacular as a reason to look for an error first.

Invalidation

If you find yourself changing the rule after seeing the results, adding a filter for each losing period or choosing the test dates to make it look better, the research has turned into curve fitting. Stop and start again from the written rule.

Common mistakes

  • Expecting a quantitative approach to remove losses. It controls the size of losses, not their existence.
  • Treating a backtest as a forecast. It is a description of the past under stated assumptions.
  • Competing where an individual is weakest: very short holding periods, illiquid stocks, strategies whose profit is smaller than the trading costs.
  • Abandoning the rules during the first losing streak, then adopting new ones after a winning streak someone else had.
  • Treating “No edge” as a failure to hide rather than a result to record.

Summary

Quantitative trading means writing your decisions down as rules, testing them fairly on past data and only then trading them. Rules make a method testable, consistent and measurable; they do not predict the future, prevent losing streaks or stop edges from fading. An individual trader is best placed with daily data and holding periods of days to months. Expect most ideas to fail, expect good ones to be modest, and judge yourself by the honesty of the process. The next lesson turns a trading idea into a rule set that can be tested. For the basics of backtesting in general, see the backtesting article; for the parts of a complete system, see trading systems.

For educational purposes only. Not investment advice.

Check your understanding

Answer each question to see the explanation. Answers are not saved.

  1. Which of these is a rule that two people would always apply the same way?

    1. A Buy quality growth companies whenever they look temporarily cheap
    2. B Buy at the next open after a close above the 200-day SMA
    3. C Sell a position once its momentum starts to fade and the chart looks heavy
    4. D Add to winners when the market mood feels supportive
    Show the answer

    Answer: B. A rule uses exact, measurable conditions known at the time of the decision. The other options depend on judgement, so they cannot be tested honestly.

  2. A rule-based trader has a losing quarter. What can a written rule set plus a trade journal tell them that discretion usually cannot?

    1. A Whether the rule failed or the rule was not followed
    2. B Which of the stocks it holds are likely to recover in the next quarter
    3. C That the losses were caused by bad luck alone
    Show the answer

    Answer: A. With written rules, each trade can be checked against the rule, so execution errors and rule failures can be told apart. Rules say nothing about future prices.

  3. Where does an individual trader with end-of-day data compete most realistically?

    1. A Intraday scalping, where order speed and co-location decide most of the profit
    2. B Rules held for days to months on liquid stocks and ETFs
    3. C Thinly traded small caps, where few professionals look
    4. D Strategies that earn less per trade than the spread
    Show the answer

    Answer: B. Very short horizons reward speed and low execution costs that individuals lack. Daily rules on liquid instruments are where those disadvantages matter least.

  4. A first backtest shows three times the market’s return with a maximum drawdown of 4%. What is the best first response?

    1. A Start trading at full size at once, before other traders find the same edge
    2. B Search for an error such as look-ahead, missing costs or survivors only
    3. C Add more filters to push the drawdown lower still
    Show the answer

    Answer: B. Spectacular results are far more often a mistake than a discovery. Check the data, costs and fills before anything else.

  5. A fair test of your idea ends with the verdict No edge. What is the most useful way to read it?

    1. A The test must have been set up wrongly, so rerun it with new settings until it passes
    2. B The rule should be traded anyway, with a smaller stop
    3. C It did not beat holding the same stocks; record it and move on
    4. D No edge means the market is too random to trade
    Show the answer

    Answer: C. No edge is a common and informative result: it saves the money you would have lost learning the same thing live. Recording it keeps your research honest.

  6. Why might a classic rule from a well-known book work less well today than in the original study?

    1. A Once a pattern is widely known, more traders act on it and its reward tends to shrink
    2. B Old rules only work on monthly charts
    3. C Published rules are always false
    Show the answer

    Answer: A. Research by McLean and Pontiff found that returns of published anomalies were lower after publication. Treat any published rule as a starting point to test, not a promise.

Continue in Clear Tape

With a free account, your progress and quiz scores are saved, and you can practise with the question bank, keep a wrong-answer book and study real chart examples.