The Cost of Stepping Out

What risk management actually costs, and why it depends on what you own

An Advising Alpha research paper. Everything in this study is backtested: rules applied mechanically to historical market data, with the methodology and replication steps published in full at the end. Backtested results are not live returns, and past results do not guarantee future ones.

Every investor believes the same trade-off: protection costs performance. Get defensive and you give up return. Stay fully invested and you accept the crashes as the price of the compounding. Pick your poison.

We spent the past month testing that belief against 55 years of market history, and it broke in a way we did not expect. The cost of protection is not what everyone thinks it is. Sometimes it is close to zero. Sometimes it is enormous. And the thing that decides which, almost by itself, is what you own when the signal tells you to leave.

This paper walks through what we found in three parts. First, the surprise: on a plain index, stepping out has historically cost almost nothing. Second, the price tag: on portfolios built to beat the market, the same discipline gets expensive, and we can say exactly why. Third, the part that decides everything anyway: the human holding the account.

One promise before we start. Every number below came out of the same testing engine, run on public data, with the rules written down before the results were seen. The recipes are in the appendix. If you want to check our work, you can.

Part one: the free lunch nobody talks about

Here is the experiment. Take a simple signal, a light that is either green or red, computed once a day from public data after the close. Green means own the S&P 500 through an index fund. Red means sell and sit in Treasury bills until it turns green. When the light changes, you trade the next day, because you cannot act on a close you have not seen yet. Cash earns real T-bill interest while you wait.

Now add a dial. The dial sets how much stays invested when the light is red. At 100, you ignore the light completely: pure buy and hold. At zero, you follow it completely. Every setting in between is a blend, and here is a useful piece of arithmetic: holding half your money as permanent buy and hold and half fully signal-managed is mathematically the same portfolio as setting the dial to 50. The wealthiest families run this as “core and satellite.” It is the same machine.

We ran that dial across 55 years, 1971 through mid-2026, using a two-part signal we will detail in the appendix: the S&P 500’s position against its 200-day average, confirmed by the breadth of the market itself. Here is what $10,000 would have become at each setting.

Backtested, 1971 to July 2026. Trades at the next day’s close after each signal change; cash earns 13-week Treasury bill interest; dividends included from 1988 onward, price-only (scaled) before, which understates the early years equally at every setting. Not live results.

Read the two middle columns separately, because the whole finding lives in the gap between them. The growth column is a tie. Across the entire dial, from full protection to none, the spread is two tenths of a percent a year, which over 55 years is statistical noise. Nobody in that table got rich from the timing itself.

The worst-fall column is not a tie. It swings 32 points. The buy-and-hold investor watched more than half their money vanish twice, in 1973-74 and again in 2008-09, and their single worst year took 48% of the account. The investor who followed the light fully never fell more than 23% and never had a year worse than 19%.

Same destination. Wildly different ride. Historically, the market did not charge for the choice.

That is the sentence that should stop you, because it contradicts what every finance textbook implies. Protection was not expensive. Across 55 years, on the index, choosing your worst day was roughly free, and the only real question was which worst day a given human could survive. Hold that thought, because part three is about that human.

First, though, an honest complication. The free lunch was not evenly spread across time.

When the insurance paid, and when it charged

We think of long market history in eras: roughly 1970 to 1982, a grinding sideways stretch. 1982 to 2000, a historic expansion. 2000 to 2013, the lost decade and its aftermath. 2013 to now, another long climb. Split any timing rule along those lines and a pattern appears that we could not make go away, because it shows up in every trend signal we tested.

Take the simplest one, 12-month momentum: at each month’s end, is the index higher than it was a year ago? In the 1970-1982 stretch, following it would have earned 6.9% a year against 3.5% for holding, while cutting the worst fall from 48% to 20%. In 2000-2013, it would have earned 5.7% against 2.1%, with a worst fall of 19% against 55%. Those are enormous wins. But in 1982-2000 it would have trailed buy and hold by over a point a year, and since 2013 it has trailed by three and a half.

So the honest statement has two clauses. Risk management is insurance that has paid for itself, handsomely, in the long sideways eras. And it has charged a real premium during the long expansions. Anyone who tells you one side of that sentence without the other is selling something.

Which raises the obvious question: if the current era is an expansion, why carry the insurance at all? Two reasons. Nobody rings a bell when an era ends, and the people who carried no insurance into 2000 and 2008 did not get to renegotiate. And second, as part one showed, over full history the premium netted out to roughly nothing. You were not paying for the insurance. You were paying with volatility in the years it was not needed, and being repaid in the years it was.

Part two: the price tag appears

Everything above was tested on the index. But most serious investors do not hold the index. They hold portfolios built to beat it. So we ran the same experiment on our own five model portfolios, applying the same lights to concentrated stock portfolios with long backtested records. This is where the free lunch ended, and where the most useful principle in this paper showed up.

The short version: the same overlay that was free on the S&P 500 behaved completely differently depending on which portfolio it was protecting.

On our flagship Core 20, a diversified 20-stock portfolio, the breadth-based overlay would have added return, 13.9% a year against 13.1% held, while cutting the worst fall from 44% to 18%. Better return and less than half the pain. On Market Masters, similar: 17.2% against 16.6%, with the worst fall dropping from 56% to 32%.

On Apex Momentum, our momentum-ranked portfolio, every overlay we tried cost return, one to seven points a year. And on BioTech 10, our sector portfolio, the results were genuinely bad: every market-based signal reduced return and made the drawdowns worse, in some versions much worse. The signals were green while biotech crashed through 2021 and 2022, and red while it recovered, because biotech marches to its own cycle and a market signal cannot hear that drummer.

Why such different answers from the same discipline? Three mechanisms, and they generalize far beyond our portfolios.

First: the cost of stepping out scales with the alpha of what you step out of. An index fund has no edge to forfeit; when you sit in T-bills, you miss the market’s return and nothing else. But a portfolio that beats its benchmark by several points a year loses those points too, every month it sits out. BioTech 10’s backtest beats its own sector by roughly six points a year, and the sector-signal version of its overlay sat out 106 of 291 months. That is a decade of forfeited edge. The more alpha you own, the more every month on the sidelines costs, which means risk management is cheapest exactly where most people apply it least: on plain index exposure.

Second: do not stack two timers. Apex Momentum already times itself; ranking by momentum quietly walks the portfolio out of weakening names every quarter. Adding an in-or-out market signal on top doubles the same bet, and in an expansion era, doubling it just doubles the drag.

Third: match the signal to the asset.When we replaced the market signal on BioTech 10 with a biotech-sector signal, a 20-day versus 200-day average on the Nasdaq Biotech index, the overlay finally worked in the right direction: it was the first version that actually reduced BioTech’s worst fall. And at a partial dial setting, keeping 75% invested on red, it would have matched the held return of 16.3% a year while cutting the worst fall from 71% to 61%, ten points of relief at no historical cost. A sector portfolio needs a sector signal, and even then, position sizing does more of the work than any signal can.

All model-portfolio figures backtested, monthly data, signal state at each month’s end applied to the following month; details and windows in the appendix. Not live results.

And the era pattern from part one reappeared here, right on schedule. Core 20’s overlay earned its entire keep in the 2001-2013 consolidation, where it would have added three points a year while cutting the worst fall from 44% to 14%. Since 2013, the same overlay would have cost almost two points a year. The insurance pays in the grinding eras and charges in the climbing ones, whether the vehicle is an index or a stock portfolio. The pattern would not go away because, we think, it is the real structure of the thing.

Part three: the human holding the account

Everything so far assumed a robot: a holder who executes every signal the day after it fires, without hesitation, forever. We have never met that robot. So the last part of this research looked at what these systems demand from an actual person, and this is where the paper stops being about signals at all.

Start with the workload. The gentlest signal we tested changed its mind 36 times in 55 years, less than once a year. The most protective one changed 340 times, roughly six times a year, and one of the confirmation-style combinations changed 613 times. Every one of those changes is a decision delivered to a human at an emotionally inconvenient moment, because trend signals only fire when the market is already moving. Six decisions a year for 55 years is 330 chances to hesitate, to second-guess, to wait one more day. Nobody executes 330 uncomfortable decisions cleanly. The systems with the best drawdown numbers demand the most from exactly the machinery, human attention and nerve, that fails first.

Next, the blind spot, and this one is structural. Slow signals dodge slow bears and miss fast ones entirely. Monthly momentum would have sidestepped 1973-74 almost completely, down 3% while the market fell 48%, and sidestepped 2008-09, down 9% against 55%. Those are the wins that built its record. But in 1987 it rode down 31 of the market’s 33 points, and in 2020 it absorbed the entire 34% crash, because a light that checks once a month against a year-old anchor cannot react to a five-week collapse. No amount of parameter tuning fixes this. It is what slowness costs, purchased in exchange for calm. You should know which trade you are making.

Then there is the failure mode that does not show up in any backtest, because backtests cannot make it: acting on a signal that is not finished yet. Several traders we know follow a weekly market-breadth gauge, the percentage of S&P stocks above their 50-day average. In April of 2025, mid-week, that gauge printed 9.8, below the classic bear trigger of 10. Anyone watching intraday saw it with their own eyes. But weekly charts repaint until the week closes, and when that week locked, the final reading was 10.3. No trigger. The chart’s official history says the signal never fired. Anyone who acted mid-week sold into what became one of the sharpest recoveries of the decade, on a signal that, by the only record that counts, never existed. The market then gaslights you: you go back to the chart, and the moment you acted on is not there.

That is not a story about one bad indicator. It is the general lesson of this whole section: every moving part you add is a decision you will someday have to make at the worst possible moment, under time pressure, with money and reputation attached, and the historical record of humans making decisions under those conditions is poor. We built a testing engine precisely because we do not trust ourselves to eyeball this stuff, and the engine’s most consistent finding is that the gap between a system’s backtest and a human’s result is made of exactly these moments.

Which brings the three parts together into one conclusion we did not fully expect when we started.

The data says the return difference between protection and no protection was, on the index, close to zero. The data says the cost appears when you protect high-alpha portfolios with the wrong signals, or stack timers, or trade six times a year. And the data says the worst falls of buy and hold, 55% with a 48% worst year, are precisely the magnitude of loss that, in our experience, almost no investor holds through. They sell somewhere near the bottom, sit out the recovery, and re-enter after it feels safe again, which converts a temporary 55% drawdown into a permanent one.

What we did about it

You can always build a better system on paper. Another indicator, another threshold, another basis point of backtested edge. We know, because we spent a month building and testing them, and the full results, including everything that failed, are summarized above.

We built Advising Alpha the other way around, and this research is why.

Our model portfolios rebalance four times a year, on a published schedule, for two deliberate reasons. The schedule follows the quarterly disclosures that map where institutional money is moving. And it means positions are bought after companies have shown their earnings, then held through at least one full earnings season, because we want to own businesses that prove themselves quarter after quarter, not trade around the proving. When a portfolio changes, Pro members get the trade list by email, executable in about fifteen minutes at any broker. Four decisions a year, delivered, with nothing to watch in between. Not because more sophistication would not backtest better. Because a system you follow beats a system you abandon, and four calm decisions a year is a system a human being can actually follow for the decades that compounding requires.

The research above is what we removed, so you would not have to carry it.

If this is the kind of work you want behind your portfolio, two ways to get more of it. The Sunday Edge is our free weekly email: where the market sits against history since 1950, one stock worth understanding, one principle to keep you disciplined. And Advising Alpha Pro opens every holding in every model portfolio, the reasoning behind each one, and the quarterly trade alerts, for $349 a year with a 14-day money-back guarantee.

Either way, the next rebalance is on the calendar. That is rather the point.

Appendix: methodology and replication

Data.S&P 500 daily, 1969 to July 2026: total return index from January 1988 onward, price-only index scaled to join smoothly before that (pre-1988 returns are therefore understated by roughly the dividend yield, equally across every strategy tested). Cash: 13-week Treasury bill yield, daily. Breadth: the NYSE cumulative advance-decline line, daily from 1970. Model portfolios: month-end backtested values as published on this site, various inception dates from 2001 to 2010.

Timing discipline. Signals are computed only from completed bars: a daily signal from the completed daily close, a monthly signal only on the last trading day of the month. Trades execute the next trading day (we tested both next open and next close; the tables above use next close, the stricter assumption). No signal is ever read intraday or intraweek. Part three explains what happens to people who do.

The dial signal (part one).Two conditions, checked daily: the S&P 500 close above its 200-day simple moving average, and the 50-day exponential moving average of the NYSE advance-decline line above its 200-day exponential moving average. Both true: fully invested. One true: exposure halfway between full and the dial’s floor. Neither: exposure at the floor. Exposure tiers were fixed before results were computed.

Other signals cited. 12-month momentum: month-end close versus the close 12 months prior. 200-day trend: daily close versus its 200-day simple average. Monthly cross: 3-month versus 10-month exponential averages of month-end closes. Sector signal for BioTech 10: 20-day versus 200-day simple average of the Nasdaq Biotech index. Each was also run across a neighborhood of nearby parameters; no result cited here depends on a single magic number.

Honesty notes. Everything here is backtested and would have applied only to someone following each rule mechanically; real investors incur taxes, costs, and hesitation the tables do not show. We tested many indicators; several will always look good by chance, which is why we report failures alongside wins and re-run everything from committed data. A rule that worked for 55 years can stop working. We publish the recipes so you can hold us to them.

Appendix: shifting our own rebalance dates

Our model portfolios rebalance on a published quarterly schedule, so as a final check we re-ran the Apex construction with every rebalance date shifted by one, two, and three months in each direction, rosters at each shifted date following the same information rule as the published schedule. All seven schedules would have beaten the S&P 500 over the window, and the fair expectation going forward is the average across the shifts, not the best row. Most of that edge comes from the feeder universe itself, shown in the Hold All column, with the momentum ranking adding 2.3 points a year on average.

Advising Alpha publishes investment research under the publisher exemption recognized by Section 202(a)(11)(D) of the Investment Advisers Act of 1940 (Lowe v. SEC, 472 U.S. 181, 1985). We are not a registered investment adviser. Educational and informational only, not investment advice. Model portfolios and performance shown are hypothetical, backtested applications of our methodology to historical data. Members who execute the same trades may not achieve the same results due to timing, fees, taxes, and individual circumstances. Past performance does not guarantee future results.