I put a stop-loss on 6 million trades.
It ate 78% of the return.
"Picking stocks is luck; cutting losses is skill." Everyone who has traded for a while has heard it. So I took stock-picking out entirely: buy any stock on any day, and just add a stop-loss. Every Korean stock, twelve years, 6.1 million trades.
01The rules
Korea's KOSPI (main board) and KOSDAQ (secondary board) trade 09:00–15:30 Korea Standard Time (UTC+9), with a daily price limit of ±30%.
- Buy every stock at the open of every trading day (= any stock, any day: 6.1 million entries)
- Hold for up to one month (20 trading days); if nothing happens, sell at the open then
- Deduct 0.22% on the buy for commission and tax
- Stop-loss s% — if the close is more than s% below the entry, sell at the next open (3, 5, 10%)
- Take-profit t% — if the close is more than t% above the entry, sell (5, 10%)
- Trailing stop s% — if the close is s% below the highest close since entry, sell (5, 10%)
- Stop + take-profit — whichever comes first
"Judge on the close, execute at the next open" is what someone who does not watch the screen all day can actually do. If the stock gaps through the stop, it sells below the stop level — as in real life.
Besides win rate, average and median, the tables show the 5th percentile (one trade in twenty does worse than this), the trigger rate, the average holding period, and a "per month of capital" figure (average ÷ holding days × 20) — because a stop shortens the holding period, and freed capital can be reused. The median is the middle result when every trade is lined up from worst to best.
02Adding a stop-loss
| Rule | Win rate | Average | Median | 5th pct | 1st pct | Triggered | Avg days | Per month of capital |
|---|---|---|---|---|---|---|---|---|
| No rule | 44.7% | +0.67% | −0.88% | −19.9% | −32.7% | — | 20.0 | +0.67% |
| Stop 3% | 26.9% | +0.15% | −3.36% | −8.4% | −14.1% | 68.0% | 10.3 | +0.29% |
| Stop 5% | 33.7% | +0.20% | −4.60% | −10.3% | −16.0% | 54.8% | 13.1 | +0.30% |
| Stop 10% | 41.5% | +0.27% | −1.77% | −14.5% | −20.3% | 30.2% | 17.0 | +0.32% |
| Trailing 5% | 35.2% | +0.08% | −2.67% | −9.4% | −14.6% | 80.0% | 10.4 | +0.16% |
| Trailing 10% | 40.7% | +0.21% | −1.93% | −13.7% | −19.1% | 46.5% | 15.7 | +0.27% |
First, what the stop delivered. With no rule, one trade in twenty ends worse than −19.9% and one in a hundred worse than −32.7%. With a 3% stop those become −8.4% and −14.1%. The tail is cut in half. That is the product a stop-loss sells.
Then the price. The average return fell from +0.67% to +0.15% — 78% gone. The win rate dropped from 45% to 27%, the median from −0.88% to −3.36%. The 3% stop fired on 68% of trades: buy any stock and two times in three it dips 3% within a month, and most of those dips were dips that came back. A stop does not wait for them.
"But the money comes back faster, so I can buy again." Per month of capital: no rule +0.67%, 3% stop +0.29%. Even with the holding period halved, it stays more than half behind.
03Take-profits, and both together
| Rule | Win rate | Average | Median | 5th pct | Triggered | Avg days | Per month of capital |
|---|---|---|---|---|---|---|---|
| No rule | 44.7% | +0.67% | −0.88% | −19.9% | — | 20.0 | +0.67% |
| Take-profit 5% | 55.4% | +0.40% | +2.19% | −18.6% | 47.0% | 14.2 | +0.56% |
| Take-profit 10% | 48.3% | +0.59% | −0.22% | −19.4% | 28.3% | 17.1 | +0.69% |
| Stop 5% + TP 10% | 37.1% | +0.12% | −4.20% | −10.0% | 75.0% | 10.7 | +0.21% |
| Stop 5% + TP 5% | 43.9% | −0.00% | −2.29% | −9.7% | 85.8% | 8.5 | −0.01% |
A 5% take-profit has the highest win rate in the article (55.4%) and the only positive median (+2.2%). But its average, +0.40%, is below no rule at all: it sells the winners at 5% and misses the ones that would have run. The tail (−18.6%) is untouched. It buys win rate and pays with average.
The most popular combination, 5% stop + 5% take-profit, has an average of exactly zero. 86% of trades hit one side or the other within 8.5 days, winners and losers cancel, and only the cost is left. Stop 5% + take-profit 10% is +0.12%, effectively the same.
Rules change the shape of the distribution: a stop trims the left tail, a take-profit trims the right one.
Not one rule raised the average.
04Three months, and large caps
With a three-month maximum hold the picture sharpens. No rule averages +2.21%; a 3% stop +0.51%; a 5% stop +0.64%. The 3% stop fires on 82% of trades and wins 16.9% of the time — almost no stock goes three months without a 3% dip, so a 3% stop is nearly "stop out everything". The 5th percentile improves from −30.5% to −8.9%. The ratio of what is bought to what is paid is the same as at one month.
Large caps only (market cap above 100 billion won, about US$70 million): no rule +1.37% at one month, 3% stop +0.51%, 5% stop +0.66%, 10% stop +0.89%. Large caps start from a higher average, so the stop's share falls to 63%, but the direction is identical and the tail again halves, −19.1% to −8.3%.
05So are stop-losses wrong?
No. The question was "does a stop-loss raise returns?" and the answer is no. A stop-loss is not a return tool; it is insurance. Insurance does not raise expected value. It caps the worst case in exchange for a premium. Here the premium was 60–80% of the average return, and the payout was half the tail.
Some people need that insurance: anyone concentrated in one stock, anyone on borrowed money, anyone for whom a −30% hit means no next trade. For them −32.7% is not an expected-value question but a survival one, and turning it into −14.1% is worth 78% of the return.
For someone spread across many stocks who can absorb a single −30%, the stop was a device for getting out of a dip that was going to recover, two times in three. "Cutting losses is skill" reads, in this data, as "buying insurance is skill" — skill only for those who need it.
06Caveats
This is any stock, any day. With a selection rule the value of a stop could differ. But the claim that "a stop alone improves things regardless of selection" is rejected here.
Judged on the close, executed at the next open. Intraday stops or resting stop orders differ slightly, but selling worse on a gap-down happens either way.
Costs are the 0.22% buy-side deduction only. Stop rules double the number of trades, so the real cost gap is larger than the tables show. Delisted stocks are not included — with them, the no-rule tail would be worse.
07What I take from it
- A 3% stop halved the worst losses (5th percentile −19.9% → −8.4%)
- It took 78% of the average return in exchange (+0.67% → +0.15%); 5% and 10% stops took 60–70%
- The 3% stop fired two times in three, mostly on dips that would have recovered
- Take-profits raise win rate and lower the average. Stop + take-profit averages zero
- No rule raised the average. A stop-loss is insurance, not a return tool
The question to ask before drawing a stop line is not "at what percent" but "am I someone who needs insurance?" If everything is in one stock, yes. If it is spread out, then in this data the stop line was peace of mind bought with two-thirds of the return.
댓글 쓰기