Build and test a strategy Lesson 2 of 14
Strategies · Building and testing a strategy
Systematic vs discretionary trading: rules or judgment
Compare rule-based and judgment-based trading, and see on 33 years of SPY what happens to a simple rule when its follower hesitates, skips signals or overrides it.
On this page
Every decision to buy or sell is made in one of two ways. A systematic trader writes the decision down in advance as a rule and then carries it out. A discretionary trader looks at the situation each time and decides. Most people do some of each, usually without noticing where one ends and the other begins.
The difference matters for a practical reason. A rule can be tested on the past before any money depends on it. A judgment cannot. This lesson defines the two approaches, sets out what each is good and bad at, and then does something a discussion of this topic rarely does: it takes one simple rule, tested on 33 years of SPY, and measures what happened to it when the person following it bent it in three ordinary, human ways.
Two ways of deciding
Systematic trading means the rules exist before the trade. They say what to buy or sell, when, how much, and when to get out. A rule is complete when it leaves nothing to decide at the moment of action. "Hold the fund when its month-end close is above its 10-month average, otherwise hold cash" is complete. "Buy strong stocks on pullbacks" is not, because strong, pullback and buy all still need a decision.
Systematic does not mean automated. A person who checks one number on the last day of each month and places one order is trading systematically with a pencil. It also does not mean complicated, or fast, or free of judgment altogether. Choosing which rule to follow, on which market, with how much money, is a judgment. The system only begins after those choices.
Discretionary trading means a person weighs the situation and decides each time. The inputs may be the same charts and numbers a system would use, plus anything else: news, the tone of an earnings call, a sense that this time is unlike the last. Warren Buffett buying a business after reading its accounts is discretionary. So is a trader who sells because the market "feels heavy".
Most real practice is a blend. A common one is rules for the entry and judgment for the exit, or rules for everything except the decision to stand aside before a big news event. The blend is where the trouble usually starts, because a person rarely records which decisions were the rule's and which were their own.
What each is good at
| Question | Systematic | Discretionary |
|---|---|---|
| Can it be tested on the past? | Yes, exactly as it will be traded | No; only the person's record exists |
| Is it consistent? | The same input always gives the same decision | Varies with mood, fatigue and the last result |
| Can it use new kinds of information? | Only what the rule already names | Yes, anything the person notices |
| Can it adapt to a changed market? | Not until someone rewrites it | Yes, and also when nothing has changed |
| What does it demand? | Patience through losing runs the test already showed | Skill that is real, and honest record-keeping |
| How does it usually fail? | The rule was fitted to the past, or stops working | Decisions drift with emotion; results cannot be explained |
The case for rules
The strongest evidence comes from outside finance. In 1954 the psychologist Paul Meehl compared the predictions of trained experts with those of simple formulas using the same information, in fields such as success in college, the outcome of psychiatric treatment and parole decisions. The formulas did as well or better in nearly every comparison. A later review of 136 such studies by Grove and colleagues reached the same conclusion. Daniel Kahneman's summary of this work is that people are inconsistent: given the same case twice, experts often give different answers, and a formula never does.
Markets add a second problem, which is that outcomes are noisy and feedback is misleading. A bad decision often makes money and a good one often loses. A discretionary trader learning from results is learning partly from luck. In their study of 66,465 US households with discount brokerage accounts from 1991 to 1996, Barber and Odean found that the households that traded most earned 11.4% a year while the market returned 17.9%. Those trades were judgment calls, and on average they cost money. The library's lessons on Overconfidence and overtrading: the research on trading too much and Loss aversion: why losses hurt twice as much as gains please cover why.
The case for judgment
Rules know only what they were told. A moving average rule cannot know that a company has been accused of fraud, that a market has closed, or that the instrument it trades is about to be delisted. A person can. Rules are also built from the past, and when the conditions that produced the past result change, a rule goes on trading as if they had not.
There are also decisions no rule can make: whether to trust a backtest, how much of one's savings a strategy deserves, whether to stop. Good discretionary investors exist, and the best of them tend to work in areas where information is hard to put into numbers and time horizons are long.
What bending a rule costs: a test
A backtest reports what a rule earned when followed without exception. People do not follow rules without exception. The interesting question is how much the exceptions cost.
The rule is the 10-month moving average rule from Moving average trend following: the 10-month rule, tested, on SPY.
- Data. SPY month-end closes, February 1993 to October 2026. The test runs from November 1993: 396 months.
- Trading and costs. Trades at the month-end close of the signal, 0.10% per buy and per sell.
- Left out. Dividends, interest on cash and taxes.
Followed exactly, the rule made 25 round trips, turned $1 into $11.98 and returned 7.8% a year, with a deepest fall of 23.8%. Buying and holding returned 8.9% with a deepest fall of 52.2%.
Now three followers who each change one thing. Each change is written as a rule too, so it can be tested; a real person's version would be less regular and so harder to measure.
- The hesitant buyer. After every sell signal, this person does not trust the next buy signal until it has held for two more month-ends. Having just been through a fall, they want to be sure.
- The once-bitten buyer. After any trade that lost money, this person sits out the next buy signal entirely and waits for the one after.
- The reluctant seller. This person ignores a sell signal unless the close is more than 5% below the average. Small dips under the line "do not count".
| Follower | Return per year | $1 became | Deepest fall | Round trips | Time invested |
|---|---|---|---|---|---|
| Follows the rule exactly | 7.8% | $11.98 | −23.8% | 25 | 76% |
| Buys back two months late | 6.1% | $6.97 | −17.6% | 19 | 65% |
| Skips the buy after a losing trade | 6.5% | $8.05 | −18.6% | 19 | 60% |
| Sells only 5% under the average | 8.6% | $15.01 | −25.3% | 12 | 84% |
| Buys and holds | 8.9% | $16.31 | −52.2% | 1 | 100% |
Hesitation is expensive
Waiting two extra months to buy back cost 1.7 points a year. Over 33 years that is the difference between $11.98 and $6.97, a loss of 42% of the final amount, from a habit that feels like prudence. The rule's gains come from being in the market early in a recovery. In June 2009 the exact rule bought SPY at $91.95; the hesitant buyer got in two months later at a higher price, and was late in the same way after every other sell.
Waiting only one extra month cost less: 7.5% a year and $10.67. The damage grows with the delay.
One loss should not cancel the next trade
The once-bitten buyer skipped six buy signals. Three of them turned out to be losing trades, so half the time the caution looked wise: it avoided losses of 4.5% in 1994, 1.4% in 2019 and 6.4% in 2022. The other three were the trades from April 2003 to July 2004, which gained 20.4%, from September 2010 to August 2011, 6.9%, and from March 2016 to October 2018, 31.4%.
That is the general shape of the problem. A trend-following rule loses small amounts often and makes its money on a few long trades. Whether the last trade won or lost says nothing about the next one, but it says a great deal about how the follower feels. Skipping after a loss removes trades at random, and since the winners are much larger than the losers, removing trades at random costs money. This follower also spent 72 months below an earlier peak, against 37 for the exact rule.
Both hesitant variants had a smaller deepest fall than the exact rule, because they were in the market less. Less exposure, not better decisions, produced that.
The override that helped
The reluctant seller did better than the rule: 8.6% a year against 7.8%, with 12 round trips where the rule made 25. Ignoring small breaks below the average avoided many of the quick sell-and-buy-back round trips of the years after 2009, at the cost of a slightly deeper worst fall, 25.3%.
It would be easy to conclude that discretion wins after all. Three cautions apply.
- It is not discretion. "Sell only when the close is more than 5% below the average" is a rule. It can be written down, so it can be tested, which is how this number exists at all.
- It was found by looking back. The 5% band was chosen after seeing that small breaks had mostly reversed since 2009. Split the period and its edge over the exact rule is entirely in the second half: 7.9% a year against 8.0% up to February 2009, then 9.2% against 7.7%.
- A real override is not this tidy. A person who "ignores small dips" has no fixed threshold. In a slow decline such as 2000 to 2002 the dips are small for a long time, and the reluctance that helped in 2019 keeps them in.
The fair reading of the table is narrow: the two changes driven by fear after a loss cost money, the one that amounted to a different rule happened to help in the second half of this sample, and none of the three matched simply holding on return.
The split
| Follower | Nov 1993 – Feb 2009 | Feb 2009 – Oct 2026 |
|---|---|---|
| Follows the rule exactly | 8.0% a year | 7.7% a year |
| Buys back two months late | 5.6% a year | 6.5% a year |
| Skips the buy after a losing trade | 7.0% a year | 6.2% a year |
| Sells only 5% under the average | 7.9% a year | 9.2% a year |
| Buys and holds | 3.1% a year | 14.1% a year |
The two fearful variants trailed the exact rule in both halves. That consistency is worth more than the size of the gap.
Where each approach fails
Systematic strategies fail when the rule was fitted to the past and had no real basis, when the market changes in a way the rule cannot see, and most often when the follower stops following. A rule that trailed buy and hold for most of 17 years, as this one has since 2009, is hard to keep obeying. Abandoning a system during its worst stretch and returning after its best turns its results into something worse than either the system or doing nothing.
Discretionary trading fails when it cannot be told apart from luck. Without rules there is no test, and without a written record there is not even a history, only a memory, and memory keeps the wins. The same flexibility that lets a person react to real news lets them react to their own last loss.
The blend fails in its own way: the rule gets the blame for results the overrides produced. The practical defense is a journal that marks each trade as "rule" or "override", so the two can be scored separately after a year (Trading plan and trading journal: what to write down).
How Finabase does it
Finabase's research tools are systematic by construction, and they leave the judgment to you.
- Strategy takes rules in exactly the complete form described above: a condition comparing two values, a branch to hold if it is true and another if it is false. The sample is "if SPY's 50-day SMA is above its price". It replays the rules over two years from an amount you set. Writing a rule there is a quick check on whether an idea is complete: if it cannot be entered as a condition, it is still a judgment.
- Backtest trades every signal of the chosen indicators on one symbol, $10,000 per signal, with no trading cost. It cannot skip a signal or hesitate, so its result is the fully systematic one. Neither tool resembles the test on this page in length or costs: two years for Strategy, a fixed dollar size and no costs for Backtest (Backtest indicator signals and rule-based strategies on Finabase).
- Paper trading is where the difference between the rule and the person shows up. Following a rule by hand with virtual money, and noting each time a signal was skipped or delayed, measures the gap this lesson is about before it costs anything (Practise with Finabase paper trading on the web).
No Finabase tool places trades from a rule or tells you what to buy or sell.
Frequently asked questions
Is systematic trading better than discretionary trading?
Neither is better in general. Systematic rules are consistent and can be tested before use, and research in many fields finds that simple formulas match or beat expert judgment on the same information. Judgment can handle information no rule names. What the evidence does not support is switching between the two according to how the last trade went.
Is systematic trading the same as algorithmic trading?
No. Systematic means the decisions follow rules fixed in advance. Algorithmic means a computer places the orders. A monthly rule carried out by hand is systematic and not algorithmic, and a program that executes a person's discretionary orders in small pieces is the reverse.
Why is it so hard to follow a trading system?
Because a system's losing runs arrive in real time and its backtest arrived all at once. In the test here the rule lost money on 9 of 25 trades and trailed buy and hold for most of the years since 2009. Each loss invites a small exception, and the exceptions, not the rule, then decide the result.
Can a discretionary trader use backtests?
Only on the parts of the process that can be written as rules. A trader who always requires price above a moving average before buying can test that filter. The final decision, made by eye, cannot be tested, only recorded and scored afterwards.
Summary
Systematic trading fixes the rules before the trade and discretionary trading decides each time; the first can be tested on the past and the second cannot. On 33 years of SPY, a simple monthly rule returned 7.8% a year when followed exactly. Two ordinary lapses, waiting two months before trusting a buy signal and skipping the buy after a loss, reduced that to 6.1% and 6.5%, in both halves of the period. A third change, ignoring small breaks below the average, raised it to 8.6%, but that is another rule discovered in hindsight, with all of its advantage after 2009. A backtest's numbers belong to the rule followed exactly. The distance between that and what a person does is the part of a strategy that no test can supply, and a marked journal is the only way to measure it.
Check your understanding
4 questions on this lesson. Each answer comes with the reason, and nothing is scored.
Which of these is a complete systematic rule?
- Buy strong stocks when they pull back
- Hold the fund when its month-end close is above its 10-month average, otherwise hold cash
- Sell when the market feels heavy
Show the answer
B. Hold the fund when its month-end close is above its 10-month average, otherwise hold cash A rule is complete when two people following it would make the same trades. "Strong", "pull back" and "feels heavy" all still need a decision at the moment of action, so they cannot be tested on the past.
A backtest shows a rule returned 7.8% a year. A follower obeys most signals but waits two extra months before trusting each buy. Whose result is the 7.8%?
- The rule followed exactly; in the SPY test the hesitant follower earned 6.1%
- The follower, since the delays were small
- Both, because the same signals were used
Show the answer
A. The rule followed exactly; in the SPY test the hesitant follower earned 6.1% A backtest belongs to the rule with no exceptions. Buying back two months late cut the result from $11.98 to $6.97 per dollar over 33 years, because the rule earns its gains early in recoveries.
After a losing trade, one follower skips the next buy signal. Three of the six skipped trades would have lost. Why did the habit still cost money?
- Skipping signals raises trading costs
- The skipped losers were small (1.4% to 6.4%) and the skipped winners were large (6.9% to 31.4%)
- The rule stops giving signals after one is skipped
Show the answer
B. The skipped losers were small (1.4% to 6.4%) and the skipped winners were large (6.9% to 31.4%) A trend rule loses small amounts often and makes its money on a few long trades. The last result says nothing about the next trade, so skipping after a loss removes trades at random, and random removal costs more in winners than it saves in losers.
Selling only when price was 5% under the average beat the exact rule, 8.6% against 7.8%. What is the sound conclusion?
- Overriding sell signals by feel improves a rule
- It is a different rule, chosen after seeing the data, with all of its edge after 2009; it needs its own test
- The original rule was wrong and should never be used
Show the answer
B. It is a different rule, chosen after seeing the data, with all of its edge after 2009; it needs its own test A 5% band can be written down, so it is a rule, not discretion. Up to February 2009 it returned 7.9% a year against 8.0% for the exact rule; its whole advantage came from the second half, which is what a setting chosen in hindsight looks like.
Exercise
One task to do in Finabase. Nothing is graded or saved.
Open Paper trading. Signed out, it shows a sample account. Scroll to Transaction history and read each filled order: symbol, side, type, shares, fill price and date. For each one, try to answer: was this trade made by a written rule or by a judgment at the time? What one extra column would let you score the two kinds separately after a year?
Open paper tradingWhat you should find
You cannot tell. The record shows what was traded and when, never why. That is the trouble with a blend of rules and judgment: the rule gets the blame for results the overrides produced. The missing column is a mark on every trade, "rule" or "override". The paper account does not record it, so it has to live in your own journal.
Sources and further reading
- Paul E. Meehl, Clinical versus Statistical Prediction: A Theoretical Analysis and a Review of the Evidence (University of Minnesota Press, 1954)
- William M. Grove, David H. Zald, Boyd S. Lebow, Beth E. Snitz and Chad Nelson, “Clinical Versus Mechanical Prediction: A Meta-Analysis,” Psychological Assessment 12(1), 2000
- Brad M. Barber and Terrance Odean, “Trading Is Hazardous to Your Wealth: The Common Stock Investment Performance of Individual Investors,” Journal of Finance 55(2), 2000
- Daniel Kahneman, Thinking, Fast and Slow (Farrar, Straus and Giroux, 2011), chapter 21, “Intuitions vs. Formulas”
- Finabase chart data — SPY monthly bars, February 1993 to October 2026; rule and variants computed by Finabase
