The "always use a stop-loss" rule is trading dogma, but for many, it's a wealth-destroyer. Data from practitioners like Larry Connors shows that in mean-reversion strategies, adding a stop-loss—even a wide one—consistently degrades performance by selling assets exactly when their future expected returns are highest.

This article challenges the stop-loss myth using quantitative evidence and insights from the legendary Turtle Traders. We’ll explore why arbitrary price triggers often fail in "messy" markets and provide five data-backed alternatives—from volatility scaling to strategy diversification—to protect your portfolio without sacrificing your gains.

“Always use a stop-loss” is one of those trading rules that sounds so obviously correct that it rarely gets challenged. Many brokers and trading educators present stop-loss orders as the default way to control downside and protect gains.

The paradox is that a stop-loss can reduce some losses while also reducing total returns, especially in trading styles where the edge comes from holding through temporary adverse moves. In other words, a stop-loss often improves how a trade feels (you avoided watching something go further against you) while making the expectancy worse (you got taken out right before the move that actually pays your strategy).

It helps to separate two ideas that get blended together:

A stop-loss order is a specific order type: once the stop price is hit, a stop becomes a market order. The stop price is a trigger, not a guaranteed fill.
Risk management is broader: sizing, diversification, hedging, exit design, and avoiding ruin at the portfolio level.

When people say “stops protect capital,” they’re usually thinking about the clean textbook version: price hits your stop, you exit near that price, and your loss is capped. Real markets are messier. Regulators explicitly warn that execution can deviate significantly from the stop price in fast markets, and even a normal intraday fluctuation can trigger a stop and leave you sold at a price that looks terrible by the close.

Now layer in what the data says: whether stops help or hurt is not a moral issue. It depends on what creates returns in your strategy.

A useful framing comes from the academic side: Kathryn Kaminski and Andrew W. Lo show that if returns are essentially a random walk, simple stop-loss policies reduce expected return. If returns exhibit momentum (positive serial correlation), stop-loss rules can add value. The answer is conditional.

That conditionality is why “always use a stop-loss” is a myth. In many mean-reversion and short-horizon strategies, backtests frequently show the opposite: adding stop-losses makes results worse.

Stop-loss debates often stay philosophical. The more productive approach is: what does your style of trading need to be true, statistically, for a stop-loss to improve expectancy?

Two themes show up repeatedly in both practitioner research and peer-reviewed work:

In short-term mean-reversion trading, the goal is usually to buy weakness (in a context where rebounds are statistically more likely) and exit on the “bounce.” A tight stop-loss fights that design.

Larry Connors is unusually direct about this because he publishes test descriptions, not just opinions. In one example he describes as a large-sample mean-reversion test (liquid U.S. stocks, above the 200-day moving average, buy a pullback at the 10-day low, exit on a close above the 10-day moving average), he states that when they added stops from 1% all the way to 50%, performance was lower in every case, even with the very wide stop.

His former research partner Cesar Alvarez recounts a similar pattern from earlier work: they tested maximum-loss stops (5%, 7.5%, 10%) on mean-reversion systems, then widened stops progressively out to what is effectively “no stop,” and observed that performance improved as stops widened, with the best results coming from removing stops entirely for those mean-reversion strategies.

Those are practitioner claims, so you should treat them like any other backtest claim: as style-specific and implementation-sensitive. But they match what the theory predicts when a strategy’s edge is tied to short-term reversal: selling into the drawdown removes the strategy’s opportunity to earn its rebound premium.

A related point from Alvarez’s exit research is useful here: replacing price stops with simplistic time stops is not automatically better either. In one mean-reversion example, he tested “exit after N bars” logic and found results were highly sensitive; sometimes a time exit helped certain variants, but often it did not, and he ultimately emphasizes that waiting for the bounce (signal-based exits) can be more stable than forcing a time exit everywhere.

If you take only one lesson from Connors and Alvarez, make it this: in mean reversion, stop-loss design is not a universal safety belt. It often acts like an “edge delete button.”

Mean reversion is not the whole market. Trend following lives on a very different statistical foundation: it tries to ride persistent moves, accept many small losses, and let a few big winners dominate the long-run P&L.

The classic “Turtle experiment” run by Richard Dennis and William Eckhardt trained a group commonly known as the Turtle Traders. In Curtis Faith’s description of the system, the exits included a stop-loss that was volatility-based: a maximum of 2N (two average true ranges) from the entry, with position sizing tied to N so the stop distance corresponded to a defined percent risk.

So it is not accurate to say “the Turtles had no stops.” They did. What is accurate, and what matters for your stop-loss debate, is that:

This lines up with the broader empirical finding that stop-loss efficacy depends on whether the underlying process is more trend-like or more mean-reverting.

The practitioner story and the academic story meet in one place: turnover and trading costs.

In peer-reviewed work on stop-loss strategies with serial correlation, regime switching, and transaction costs, Lo and Remorov conclude that tight stop-loss strategies tend to underperform buy-and-hold in a mean-variance framework largely because they generate excessive trading costs; outperformance is possible when return serial correlation is sufficiently high.

That is basically the clean academic version of what many traders experience: a stop-loss that triggers frequently is not just “risk control,” it’s also a machine that converts noise into realized losses, plus commissions, fees, spread, and slippage.

One practical warning before you run off and optimize stop distance: if you search across many stop settings and pick the best one on historical data, you are increasing the odds you’ve fit noise. The probability-of-overfitting problem in backtests is well documented in quantitative finance research.

The word arbitrary matters. A stop-loss that is derived from the strategy’s logic (volatility, time horizon, signal decay) is different from “I always use 2%.”

Here are the three big failure modes you see when a price-triggered stop is bolted onto a strategy without being part of the edge.

A stop-loss is not a magic force field.

Regulators spell this out clearly: a stop order becomes a market order when triggered, and the stop price is not the guaranteed execution price. In a fast-moving market, the execution price can deviate significantly.

This gets brutal around:

Some brokers and exchanges describe the same mechanical issue: once triggered, a stop-market order executes at the next available prices, which can be far from the trigger in turbulent conditions.

Yes, you can use a stop-limit order to cap the worst execution, but regulators also warn that the limit price can prevent the order from executing at all. That means you traded slippage risk for “still holding the position” risk.

So the “stop always protects capital” story breaks down precisely when protection is most psychologically needed.

If your stop is set inside normal volatility, it is not a risk control tool. It is a coin-flip machine.

The SEC’s investor bulletin warns that stop orders can be triggered by short-term intraday moves, producing executions that are substantially worse than the stock’s closing price for the day.
FINRA gives the same warning in plainer language: a stop order may be triggered by a short-lived dramatic price change, and the stock may later resume trading near its prior level, leaving the investor having sold at an undesirable price.

This is how traders “bleed to death”:

If, after reading that, your reaction is “I’ll just widen the stop,” that’s a rational thought. But widening stops has a hidden consequence: you are no longer controlling loss-per-trade with the stop, you are controlling it with position sizing (whether you admit it or not).

This is the quiet killer, and it is the most relevant to pullback, reversal, and short-term quant strategies.

A lot of empirical finance research documents short-term return reversals at horizons like a week or a month, and a large literature studies why those reversals exist (liquidity, microstructure, behavioral effects, and more).

A simple mean-reversion strategy is essentially an engineered way to harvest that tendency: buy after short-term weakness in a favorable regime, then sell when the bounce occurs.

If you glue on a fixed percent stop-loss, you are systematically doing something like this:

That is exactly the “accumulating losses” dynamic Connors describes: stops can prevent some losses from getting worse, but they also convert many trades that would have reverted into booked losses, and those extra losses can outweigh the protection on the tail.

If you accept that arbitrary price stops are often a drag, the next question is obvious: what replaces them?

The key shift is that you stop thinking in terms of “saving each trade” and start thinking in terms of “keeping the portfolio alive long enough for my edges to play out.”

Here are five alternatives that are supported by established portfolio theory and by the way systematic managers actually design risk.

Diversification is not just about holding many tickers. It is about holding return streams that are not the same trade in disguise.

This idea goes all the way back to Harry Markowitz’s foundational formulation of portfolio selection, where the covariance between assets (and hence diversification) is central to risk reduction.

It is also the core of what Ray Dalio calls the “holy grail of investing”: finding roughly 10 to 15 good, uncorrelated return streams and balancing them so risk drops materially without necessarily sacrificing expected return.

How to apply this as a trader, not a pension fund:

The practical payoff is that diversification reduces the need to “hard stop” individual trades. If one sleeve is having a bad month, another sleeve may be in its good regime.

Even if you are mainly an equity trader, portfolios are still exposed to common shocks: growth scares, inflation surprises, liquidity crises, credit events.

The simplest hedge is “cash,” but cash is not always the best diversifier. Two widely used diversifiers are high-quality government bonds and gold, but correlations are regime-dependent and can flip.

On bonds: Vanguard summarizes a key reality: stock-bond correlation has often been negative over multi-year windows in recent decades, but it can turn positive over shorter windows, so you cannot assume bonds will always hedge equities.

On gold: the World Gold Council documents that gold’s correlation with equities varies and that gold has historically tended to become more negatively correlated during equity sell-offs, which is one reason it is used as a diversifier.

On currencies: there is also a serious literature on “safe haven currencies.” For example, Ranaldo and Söderlind document that the Swiss franc and Japanese yen tend to appreciate against the U.S. dollar when U.S. stock prices fall and volatility rises.
An IMF working paper similarly finds risk-off behavior consistent with safe haven characteristics for the yen (and to a lesser extent the Swiss franc).

Actionable takeaway: instead of relying on a stop-loss to save each trade, you can reduce portfolio fragility by combining exposures that respond differently to stress. This will not eliminate drawdowns, but it can change their shape.

If you want one “drop-in replacement” for the safety feeling of stop-losses, this is the closest.

Position sizing and volatility scaling attack the root problem: ruin is usually caused by position size, not by the absence of a stop order.

This is not just trader folklore. The portfolio research on volatility-managed and volatility-targeted approaches shows that scaling risk down when volatility is high can improve risk-adjusted performance in various contexts.

Moreira and Muir show that volatility-managed portfolios (taking less risk when volatility is high) can generate large alphas and improve Sharpe ratios across several return streams in their study.
Harvey and coauthors discuss how volatility targeting can reduce the likelihood of extreme outcomes, partly because volatility clusters and the portfolio naturally de-risks in high-vol regimes.

A very practical way to implement this in trading terms:

Notice what just happened: you controlled risk without needing the market to hit a specific price trigger. That is one reason the Turtle approach ties stop distance to N (ATR) and then sizes the position off that same volatility estimate.

Time-based exits are underrated because they feel simplistic. But they solve a real problem: once a trade has overstayed the time window where your edge exists, you are often just holding risk without getting paid for it.

In practice, many mean-reversion systems get most of their edge quickly.
Alvarez’s tests on N-day exits show that “exit after X bars” can reduce time stuck in long losers, but it is not universally additive and can be sensitive to parameters. Still, he notes traders may keep time exits because they improve the ability to stick with the system and limit extended underwater positions.

A robust way to use time exits is not to treat them as the primary exit but as a backstop:

This aligns with what regulators emphasize indirectly: stops can be triggered by noise. A time stop is not immune to noise, but it does not force you to sell precisely at the worst microstructure moment, and it avoids repeated stop-out/re-entry cycles if your signal logic is stable.

This is the fifth alternative, and it is closer to what institutions mean by “stop-loss rules” than what retail traders mean.

Kaminski and Lo define stop-loss rules as policies that reduce portfolio exposure after cumulative losses breach a threshold. Their framework makes it explicit that these rules can add or subtract value depending on the return-generating process.

This differs from a per-trade stop in three important ways:

A practical portfolio-level rule might look like:

This kind of overlay is also discussed in practitioner portfolio research as a way to manage behavioral and tail-risk problems, but with the explicit warning that protection often comes with a cost in Sharpe or returns.

Stop-losses are not “good” or “bad.” They are a tool that changes your return distribution.

The research and the real-world mechanics point to a cleaner, less dogmatic conclusion:

Stops are mainly about avoiding ruin. Diversification and sizing are about building wealth without needing every single trade to be right.