Blog & Insights

Every few weeks someone asks the same two questions. Why 20% for stocks and 15% for an index? And can I change it?

The short answers are: because a stock moves more than the market it belongs to, and no. This post is the long answer, including the part where our own data made us look harder at something we had not questioned.

Where the numbers come from

A trailing stop has to be wide enough that normal noise does not trigger it, and tight enough that it fires before a decline turns into a disaster. Where that line sits depends on how much the thing normally moves.

A single stock can lose a fifth of its value on one earnings call. An index cannot, because it is an average of dozens of them — the bad news in one is diluted by the other forty-nine. So the index threshold has to be tighter than the stock threshold. Not because we are more nervous about indices, but because the same percentage means something entirely different.

Think about what it takes for the AEX to fall 20% from its top. That is not a bad quarter for one company. That is most of the index down, with the weakest members down 40% or 50%. By the time a broad index has given up a fifth, the damage inside it is long done.

That is the whole design:

• Stocks: 20%. Individual company risk, and single stocks are volatile by nature.
• ETFs: 15%. Most of them track an index, so we treat them like one.
• Indices: 15%. An average moves less, so a smaller move means more.
• Commodities: 20%. Comparable volatility to a single stock, without the company risk.
• Crypto: 30%. A 20% week is a normal week.

These are principles, not fitted parameters. We did not search for the number that would have produced the best returns over the past two years. That number exists, and it is worthless: it describes what already happened, not what comes next.

What the data says about it

We ran every threshold from 5% to 35% across 655 assets with at least 250 trading days of history, and counted three things per threshold: how often the stop fired, how often the price was back at its old top within 90 days, and how often the price fell less than 5% further after the signal. That last one we call a false alarm — the stop fired and then nothing much happened.

For stocks the picture is exactly what you would hope for:

Tighten it to 5% and you get an alert roughly every six months per position, half of which lead nowhere and most of which bounce back. Loosen it past 20% and you keep giving up more before you hear anything, while the false alarm rate barely improves. Twenty sits where the curve stops paying you for waiting.

Crypto at 30% looks similar but cleaner: 0.44 breaches per year, and of those that broke at 30%, none were back at their old top within 90 days. When crypto breaks a 30% trailing stop, it has usually broken for a reason.

Then we looked at indices, and the number fell apart

At 15%, our index category shows 62% false alarms. Read that as a rate and it looks damning — most signals doing nothing.

It is not a rate. It is noise.

Of 49 indices, 16 had no breach at all in more than two years. Another 22 had exactly one. A percentage computed from one observation per asset is not a measurement of anything. The AEX broke once. The FTSE never. The S&P 500 once. The DAX once.

What actually produced the number was a small group of assets that do not belong in the same bucket:

• The VIX sits in our index category and accounts for four of the breaches. It does not use a trailing stop at all — it has a fixed alert level of 24. It should never have been in this comparison.
• The PHLX Semiconductor Index has a 90-day volatility of 44 and broke three times. That is not a market. It is one sector, and a famously cyclical one.
• The KOSPI and KOSPI 200 show volatility above 60, which is stock territory for a broad national index, and together account for seven breaches.

Take the VIX out and the seven most volatile members with it, and the remaining 41 broad indices produce 28 breaches across more than two years. That is roughly one breach per index per 3.5 years. There is no rate to compute there, and no reason to change the threshold.

The real question is not the percentage

It is what counts as an index.

The same problem showed up with currencies, from the other direction. We tracked 20 forex pairs. At a 20% threshold there were four breaches in 2.2 years, none recovered within 90 days — and every single one was a false alarm. Which can only mean one thing: those pairs did come back to their old top, just slower than three months. The stop fired, no further loss was prevented, and the price returned anyway.

That is the worst possible outcome for a rule like this, and it makes sense once you say it out loud. A trailing stop assumes two things: that the top means something, and that the downside is worth avoiding. A currency pair has no top it is supposed to return to — it is a ratio between two economies, drifting around whatever rates and policy put it. And EUR/USD is not going to zero. A company can.

So we removed forex from the rule. As of this week, currency pairs show a price and a top on the dashboard, and no stoploss level.

The semiconductor index raises the same question and we have not answered it yet. Should a sector index with a high volatility sit under the index threshold, or the stock one? On the data we have, it probably belongs with stocks. We would rather say that out loud than quietly change a number.

Why you cannot set your own

This is the question we get most, and the honest answer has almost nothing to do with the data.

Suppose we gave you a slider. On the day you set it, you would pick something sensible. The problem comes later, on the morning a position you believe in is sitting just above its stop, and you have a very good reason why this particular case is different.

You would widen it. Everyone would. That is not a character flaw, it is the entire reason a rule exists — to be decided once, when you are calm, and to hold when you are not.

A trailing stop that you can adjust while it is being tested is not a rule. It is a preference, and preferences lose to the moment. The value of 20% is not that it is the optimal number. It is that it was chosen before you had a position that needed it to be 25%.

What this does not tell you

Two things worth saying plainly.

Two years of history is short. Most of our assets have around 2.2 years of daily closes, which covers one meaningful drawdown and not much else. Everything above describes that period and not the next one.

And a backtest is not the system. The simulation arms a new breach only after a new top; the live system also resets the top upward when a breached asset recovers by the rule's percentage from its low. Close, but not identical. Where the two disagree, the live system is what you get.

The thresholds are in the FAQ, and the underlying data is the same breach history that feeds the analysis at the top of this page.

A trailing stoploss is a line that moves up with the price and never down. When the price touches that line, you get an email. That part has not changed, and nothing in this post changes it.

What changed is how we count what happened afterwards — and the honest version is that we had been counting it wrong.

The problem

A price does not cross a line once and walk away. It sits on it.

An asset drops through its stop at 10:15. By 10:30 it is a tenth of a percent above the line again. At 10:45 it is back below. Each of those crossings wrote a new row in our breach history. Each row had its own breach date, its own recovery, its own entry in the statistics.

To anyone holding the position, that is one bad morning. In our database it was three events.

We measured how often this happens. Of 6,883 recorded breach rows, 6,243 were a continuation of the row before it — the same asset, the same stop level, no meaningful move in between. They span 388 of the 423 assets that have ever breached. The median gap between a recovery and the next breach is one hour. The longest single chain runs to 70 rows.

It happened while we were writing this. One holding dropped through its stop at 18:33, was back above the line by 19:03, and broke it again at 19:18. Three rows, forty-five minutes, one event.

So our breach history was, for the most part, not a list of events. It was a log of price updates near a line.

What we changed

Crossings on the same stop level are now grouped into one episode, unless the price genuinely got away from the line in between.

"Genuinely" needs a number, so here it is. Between the stop and the top it was measured from, there is a distance — 20% for a stock, 30% for crypto, 15% for an ETF or index, 10% for a currency pair. A bounce has to clear 5% of that distance to start a new episode. For a stock that means climbing back to within 19% of its top. For crypto, 28.5%.

Below that, the position never really left the line, and we treat it as the same episode: the first breach price, the lowest point across the whole stretch, the final recovery, and the full elapsed time from start to finish.

We looked for a natural break in the data to put that threshold at, the way we usually do. There isn't one — the distribution of bounce sizes runs smoothly from nothing to large with no gap anywhere. So the number comes from the rule instead of from the data: an asset sitting 19% below its own top has not recovered. It is still in the situation the alert warned about.

What this does to the numbers

Our breach count drops from 6,883 to 1,051 — 85% of the rows were continuations. Breaches per asset drops from 16.3 to 2.5.

That is not a smaller dataset. It is the same data, counted once per event instead of once per price update.

It also fixes a bias we had not accounted for. Every statistic weighted by breach count was quietly weighted by twitchiness: one asset contributed seventy rows to a single stretch, while an asset that broke cleanly and kept falling contributed one. Our comparison of calm-market versus panic-market breaches was built on exactly that kind of count.

Giving each event one vote sharpened it rather than softened it. Of breaches that happened while the VIX was below 19 — a calm market, no general panic — 28% have climbed back to within 10% of the top they broke from. For breaches during a VIX above 24, it is 52%. The old counting put those at 26% and 48%. The pattern we had been describing is real, and slightly stronger than we thought: a stop broken when nothing else is going wrong is the one that tends to stay broken.

Time in alert changes too, and in the direction you would expect. Previously a three-week decline that bounced above the line twice was recorded as three short alerts. Now it is one alert of three weeks, which is what it was.

One smaller correction

Touching the stop counts as a breach. That was always the intent, but floating point does not respect intent: a stop calculated as 4496.400000000001 and a price of 4496.4 are not equal to a computer, even though they are the same number. Whether an exact touch registered came down to invisible decimals.

Both comparisons now allow for that, in the same direction. Touching the line is a breach. Being above it is a recovery. There is no longer a sliver in between where the answer depends on rounding.

What does not change for you

• Your alerts. Same rule, same trigger, same timing — real time, at the first price update after the breach.
• What arrives in your inbox. Nothing about the emails changed today. Repeat alerts for the same asset are suppressed for seven days, which already covered most of what this change merges — half of these crossings happen within an hour of each other. Where it does not line up is a position that hovers on its line for weeks: that can still produce more than one email for what we now count as a single episode. We think the mail rule should follow the same definition, and that is the next thing we will look at.
• Your positions and your plan. Nothing was recalculated, moved or reset.
• The underlying record. Every original row is still stored, untouched. The grouping happens when we read the data, not when we write it, which means the threshold can be revisited and the old counting can be reproduced exactly.

What "recovered" does and does not mean

One number deserves a warning label, and it is ours.

Most breaches recover, in the narrow sense that the price eventually climbs back above the line it broke. That does not mean the position is whole. The line sits 20% below the top it was measured from, so an asset can be "recovered" and still be a long way down.

That is why we publish both figures. Recovery says the price got back above the rule. The gap from the top says where the money actually is. They answer different questions, and only the second one is about your account.

Why we are telling you this

Because the old numbers were on our own website, and some of them are on this blog.

We could have changed the counting quietly and let the figures drift. The reason not to is the same reason the product exists: a rule is only worth anything if you apply it when it is inconvenient. We publish numbers about how often stops get broken and what happens next. If we are willing to correct those numbers only when the correction flatters us, they are not worth reading.

The revised figure is smaller and less impressive. It is also right.

We spent the past weeks building an options screener on top of the same universe we already track. We expected the hard part to be finding decent premiums. It turned out to be something else: the option that pays the most is usually the one you want least.

Start with the supply problem. On a recent run, 43 stocks passed our quality filters — uptrend intact, near their highs, above both moving averages, volatility under 30%. Of those 43, only 28 produced a single option clearing our 2% premium floor. Visa is the clearest case: at 17.5% volatility, the best call in our 20 to 60 day window paid under 1.6%. Calm, healthy companies simply do not pay much to insure.

That scarcity pushes you toward strikes closer to the current price, because that is where the premium is. And this is where the trap sits. A call 2% above the share price has roughly a 45% chance of finishing in the money. You are not being paid for time any more. You are being paid for the odds of having to sell.

Deere made this concrete for us. A call expiring in six weeks, 2.1% above the price, paid 3.4% of the share price. The same expiry at 6.5% above paid 1.8% — barely half. But the first gets assigned roughly half the time and the second maybe a quarter. The extra premium is not a better deal. It is compensation for a worse one.

Which brings us to the structure we added alongside covered calls: the Poor Man's Covered Call. Instead of owning 100 shares to write against, you buy a single deep in-the-money call with a year or more to run. For Deere that meant a call costing $110 where the shares cost $685. The call behaves almost exactly like the stock — it moves 95 cents for every dollar the shares move — but ties up a fraction of the money.

Here is the number that reframed the whole exercise for us. That same short call, written against 100 shares, returns 1.4%. Written against the long call, it returns 4.6%. The option did not change. The capital did. Which means premium expressed as a percentage of the share price — the number every options screener leads with, including ours — tells you almost nothing about what you actually earn.

The catch is specific, and it matters more on this site than most. That long call is struck about 15% below the current price. Below that level it has no intrinsic value left. Our trailing stop for stocks sits at 20%. So the position can be effectively dead before our own signal fires. If you replace shares with a long call, the stop that protects the shares no longer protects you.

Two honest caveats. Options can lose their entire value in a way shares cannot, and none of this is a reason to trade them if you do not already understand how they are priced. And everything above describes what the market paid on the days we looked. Volatility moves, and with it every number in this piece.

Our last analysis showed that two-thirds of breached assets stay well below their peak. The obvious follow-up: does it matter how nervous the market was when the breach happened? We had a hunch — that breaches during a panic bounce back when calm returns, and that the lasting damage sits somewhere quieter.

So we split every breach by the VIX at the moment it happened, and asked one question: where does that asset trade now, compared to the peak it had back then?

The pattern was clean, and stronger than expected. Of assets that breached during outright panic (VIX above 24), 73% are now back within 10% of their old peak — the typical one sits barely 1% below it. Of assets that breached while the market was calm (VIX below 19), only 17% recovered; the typical one is still about 20% under its top. The 19–24 zone falls in between.

In other words: the higher the fear when an asset broke, the more likely it has fully recovered. And the most damaging breaches happen in the quietest markets.

Why? It comes down to systematic versus idiosyncratic risk. When the VIX spikes, everything falls together — it's macro fear, not a verdict on any single company, and it mean-reverts when calm returns. But when an asset breaks its trailing stop while the broader market is relaxed, the problem isn't the market. It's that specific name. And those breaks tend to stick.

The practical takeaway flips the intuition: a breach deserves more attention when the market is calm, not less. A quiet-market breach is the market telling you something specific is wrong.

One honest caveat to close on. We only know in hindsight that this round of panic recovered. The pattern held beautifully across thousands of breaches — but the next time the VIX climbs above 24, we won't know in the moment whether it's another dip or the start of something that doesn't come back. That's the whole point of a trailing stop: it doesn't predict which one you're in. It reacts, takes you out of harm's way, and lets the recovery — if it comes — invite you back. Discipline beats forecasting precisely because no one rings a bell at the bottom.

We dug into our own numbers: 4,543 real stoploss breaches across roughly 350 stocks, crypto and commodities.

At first glance, the stoploss looks like it cries wolf. 97% of positions are back "above the line" quickly, usually within a day. It sounds like you keep selling for nothing.

But that figure is misleading. It's mostly the same nervous names crossing just below and just above their level over and over — about 13 times each on average. And "recovery" only means back above the stop, in a market that was rising anyway. That tells you little.

The real question is: where do those positions stand now, compared to their own peak? That's where it gets interesting. Two-thirds are still more than 10% below their highest point. A third are more than 20% below — dead money. The typical breached position sits about 15% under its top.

So a breach isn't a false alarm. It marks capital that has stalled. A stoploss protects you not only from losses, but above all from money sitting idle.

And the crash protection? Most of the time you don't notice it. But every so often one truly collapses — our worst case: a 96% drop. That's exactly what it's there for: cheap insurance against the rare disaster.

One more detail: breaches cluster in fear. Half happen at a VIX around 20, and 28% in outright panic (VIX above 24).

In short: a trailing stop is less about "dodging the crash" and more about not letting your money gather dust in stalled positions.