The honest answer is uncomfortable. Most signals decay. Some in weeks, some in months, some over decades. The specific timeline depends on the mechanism, the frequency, and how many people are trading the same idea. But the direction is always the same. Edges erode. Alpha becomes beta. What worked yesterday works less tomorrow.

This is not a theoretical concern. It is the central operational problem of every quantitative trading firm on the planet.

The Evidence

The most rigorous study on this is McLean and Pontiff's 2016 paper in the Journal of Finance, "Does Academic Research Destroy Stock Return Predictability?" They examined 97 published anomalies and found that returns declined by roughly 32% after a paper's sample period ended, and by 58% after publication. The mechanism is straightforward: once an edge is known, capital flows in to exploit it, and the returns compress.

That is the academic version. The practitioner version is more blunt.

Kevin Cole, CEO of Campbell & Company (a systematic multi-strategy fund running 130+ models across futures and equities), described his organizing principle on Flirting with Models podcast: "Markets are not perfectly efficient, but I do like the phrase 'efficiently inefficient.' Small edges can be found through hard research, but those edges do get competed away over time. What was once alpha becomes widely known, it gets commoditized, it becomes beta."

A colleague of Cole's put it more memorably: "Just because you uncovered dinosaur bones in an archaeological dig, it doesn't mean the dinosaurs still roam the earth."

Ernest Chan, who ran quantitative strategies for over a decade before founding Predictnow.ai, described his own experience with this on Flirting with Models: "I practically abandoned machine learning for the next 10 years in my own trading, finding simple strategies, very classical techniques, and just stick to a low-dimensional space... until a few years ago when we observed alpha decay on many simple models, including the ones that we traded. The proliferation of quant books and blogs and forums and podcasts pretty much democratized quant trading. Everybody knows about factors, momentum, mean reversion. What is the secret sauce anymore?"

Antti Ilmanen, one of AQR's most senior researchers and author of "Expected Returns," put it more diplomatically on the same podcast: "When a strategy becomes more widely known, it's reasonable to expect somehow lower Sharpe ratios. Even if you have got a solid core in your ideas, you gotta keep running to stay still. Need continuous R&D, somewhat better models, portfolio construction tools, execution skills to maintain the edge. And you should do all of this while guarding against overfitting. So that's not easy."

The Spectrum of Decay

Not all signals die at the same speed. The half-life depends on the mechanism and the frequency.

Microseconds to days (HFT). This tier requires a distinction. Some HFT signals are structural: order flow imbalance, spread dynamics, queue position mechanics. These reflect how markets physically work. They do not decay because the mechanism is not a secret to be discovered - it is a permanent feature of microstructure. Imbalance will never stop predicting short-term price moves because it is a direct measurement of supply and demand pressure.

What does decay at HFT timescales are the informational edges built on top of that structure: specific latency arbitrage routes, cross-venue stale quote patterns, or novel features that predict the next tick. These can erode in days or weeks as competitors deploy similar logic or exchanges change their matching rules. The moat for these signals is infrastructure: co-location, FPGA firmware, custom network stacks. Even a known signal remains profitable if you execute it 2 microseconds faster than everyone else. This is why firms like Citadel Securities and HRT spend hundreds of millions on infrastructure. The structural signal is table stakes. The speed and the alpha layered on top are the product.

An anonymous crypto prop trader known as Macrocephalopod described the dynamic on Flirting with Models: "You're competing with smart teams all around the world who are essentially going for the same opportunities and as a result the alpha decay is much faster than at longer horizons and the capacities tend to run lower. You really have to be thinking about execution every step of the way because it's so much of where the edge stands."

Weeks to months (stat-arb, mid-frequency). Cross-sectional equity signals, pairs trading, and event-driven strategies typically have a half-life measured in weeks to months. A novel earnings-revision signal might produce strong returns for a quarter or two before other pods at Millennium, Citadel, and Balyasny converge on similar implementations.

Giuseppe Paleologo, who served as head of enterprise risk at Millennium and head of risk management at HRT, described the dynamic on Flirting with Models: "You have an unholy turnover inside the industry of PMs that go from one platform to the other to another hedge fund. They all know each other and they talk to each other, and this generates a lot of consensus across firms." The result, he said, "is more and more resembling a game of musical chairs."

Months to years (macro, trend, short-term futures). Toby Crabel, founder of Crabel Capital Management and one of the longest-running short-term systematic traders, described this decay in vivid terms on Top Traders Unplugged. His signature strategy, opening range breakout (ORB), was a pure momentum play: if the market moved beyond a threshold from the open, you followed the direction. In the 1980s, it was one of the most profitable patterns in futures markets. "When I published the book," Crabel said, "the next 10 years was one of the best periods in history for opening range breakout." But the edge depended on a structural feature: the open was the most important reference point of the trading day. As electronic markets extended trading hours and the 24-hour session blurred the concept of an "open," the signal gradually weakened. "These changes are glacial," Crabel said. "It has changed. Yes. But Monroe Trout made a fine living with it, and so did my firm." The erosion was real but slow.

Years to decades (risk premia, carry, value, trend following). Some strategies are not really signals at all. They are risk premia, compensation for bearing an economic risk that most investors don't want. Carry, value, and trend following have persisted for a century because the mechanism is structural, not informational. Pension funds rebalance. Currencies reflect interest rate differentials. Momentum reflects the slow diffusion of information. These strategies have bad years and bad decades, but the underlying mechanism does not depend on secrecy. Rob Carver, former head of fixed income at Man AHL, put it plainly on Top Traders Unplugged: "The research cycle is quite long, so you can use the same model for years, potentially decades. If we were trading much faster, we'd be seeing very fast alpha decay and we'd need to have big research teams continuously finding new signals. The cycle is much slower." These strategies are the exception to the decay story, and they form the backbone of most long-running systematic firms.

The caveat: signals come back

Decay is not always death. Many signals go through long periods of underperformance and then return, sometimes stronger than before. Value investing looked dead from 2015 to 2020, then produced its best returns in a decade. Trend following was written off during the 2010s, then delivered spectacular performance in 2022. The Tuesday effect could reappear if market structure shifts again. A signal whose mechanism is still intact but whose returns have compressed is dormant, not extinct. The hardest problem in quantitative research is not finding signals. It is distinguishing a signal that has permanently died from one that is temporarily asleep.

What the Best Firms Actually Do

The firms that survive this problem do not try to find one immortal signal. They build systems to find, deploy, and retire signals continuously. The research pipeline is the product, not any individual model.

Russell Korgaonkar, CIO of Man Group's systematic division, described the situation on Flirting with Models: "Back then you could run a strategy and it would still work for years and years. You didn't really need to do very much. These days that shelf life's a bit shorter and therefore the value of the ongoing research process is more important. More data, more processing power, more signals, higher inefficiencies being exploited, higher bar. All of that is pointing in one direction, which means you just need to work harder to stand still effectively. Sort of that Red Queen problem: run faster and faster just to stay in place."

Korgaonkar also made a crucial distinction: "The signal research decays. The platform stuff, where you really invest in capabilities, that decay rate is much, much slower. The value of those projects can live on for years, and even sometimes decades." The signal is mortal. The infrastructure to find signals is not.

Campbell & Company runs approximately 130 individual models across four styles: trend, macro, short-term, and equity market neutral. They add 5 to 10 new models per year and remove 1 to 2. Each new model goes through a formal peer review, out-of-sample validation, and a phased deployment starting at one-tenth of target capital. Cole described their edge not as any single model but as a "meta-model": the institutional ability to generate, test, deploy, and sunset signals faster than the competition.

Crabel Capital Management shifts roughly 10% of its portfolio every year. "If we didn't do that," Crabel said, "our five or ten-year-old portfolio would have put us out of business, probably." He described the process as "vigilant" and emphasized that even small changes in market structure, on the order of 4 to 5 percentage points in win rates, can transform a strategy's character. "If that were a climate change," he said, "we would be melting the Arctic. Just a few percentage points can change so much in the way the markets behave."

Multi-manager platforms like Millennium and Citadel take a different approach. Rather than building a single research pipeline, they distribute the problem across dozens or hundreds of semi-autonomous portfolio managers (pods), each responsible for finding and maintaining their own signals. The platform provides infrastructure, risk management, and capital. If a PM's signals decay and their PnL flatlines for two to three months, they get cut.

Paleologo described the concentration: "38% of the PnL in the past three years in hedge funds was generated by three firms. The top 20 firms have generated 19% of the total PnL generated in the industry, ever." The pod model is explicitly designed for a world where individual signals are mortal but the system for finding them is not.

Why Signals Die: Four Mechanisms

1. Crowding. Other firms find the same signal. This happens through academic publication, shared talent (the "unholy turnover" Paleologo described), similar alternative data vendors, or simply parallel discovery by smart people working on the same problem. The McLean and Pontiff result is a clean measurement of this: publication alone destroys more than half the return.

Jeff Rosenberg, head of systematic fixed income at BlackRock, framed it as an arms race on Flirting with Models: "In the equity space, there's a lot more energy and attention around alphas and alpha decay minimization. That's the game in the quant equity space. It's just an arms race around new data, new techniques, alpha discovery, and then they quickly get arbed away and you need to have another one."

2. Market structure changes. Crabel's ORB decay is a textbook example. The signal did not stop working because competitors found it. It stopped working because the market structure that created the edge changed. When futures moved from pit-traded primary sessions to 24-hour electronic markets, the open stopped being the most important reference point of the day. The signal's mechanism disappeared.

3. Regime shifts. A signal trained on a low-volatility regime may fail when volatility spikes. A momentum signal calibrated to trending markets may get destroyed in a choppy, mean-reverting environment. This is not decay in the traditional sense. The signal may come back. But you need to survive the drawdown first.

4. Counterparty adaptation. If you are systematically picking off stale quotes or exploiting predictable flow patterns, the providers of that liquidity will eventually adapt. Market makers are not passive. They measure their adverse selection, identify toxic flow, and adjust their pricing accordingly. The edge does not disappear because someone else found your signal. It disappears because the person on the other side of your trade learned to defend against it.

What This Means for Smaller Traders

If you are a retail or small systematic trader, the signal decay problem might sound like a reason to give up. It is actually the opposite.

The firms I described above are managing billions of dollars. Their capacity constraints are severe. A signal that generates $500K per year is meaningless to Millennium but could be life-changing to an individual trader. Many edges that die at institutional scale remain profitable at small scale, precisely because the big firms cannot be bothered to trade them.

The practical implications are:

• Don't marry your signals. The strategy that made you money last year might not make you money next year. Monitor out-of-sample performance continuously and have a plan for when it degrades.

• Prefer structural edges over informational ones. Month-end rebalancing flows, index reconstitution effects, and seasonal patterns are driven by mechanical, recurring behavior. They persist because the source of the edge is not a secret but a structural feature of how markets operate. The Delphic Fund is built almost entirely on this principle.

• Build a pipeline, not a strategy. Even at a small scale, the discipline of continuously testing new ideas while monitoring existing ones is the difference between a temporary windfall and a sustainable practice. Campbell adds 5 to 10 models per year. You probably need 1 to 2. But zero is not an option.

• Diversify across mechanisms. If all your strategies exploit the same mechanism, they will decay together. If one is momentum-based, one is mean-reversion, and one is structural, they will decay at different rates and for different reasons. This is the closest thing to insurance against signal death.

Two Real Examples From My Own Research

Everything above might sound abstract. Here are two signals I tested on real futures data that demonstrate the decay pattern concretely.

Daily: The Tuesday Effect

Across 10 global equity index futures (ES, NQ, FDAX, NKD, and six others), Tuesday close-to-close returns captured roughly half of all weekly gains from 2008 to 2025. The full-sample Sharpe was 1.31 with a maximum drawdown of just 11.4%. Non-Tuesday days produced a Sharpe of only 0.27. The pattern was statistically overwhelming, with a t-statistic of 7.19 across the full sample.

But the decay is textbook. Annualized Tuesday returns dropped monotonically across every sub-period:

From +55% annualized in 2008-2011 (t=5.0) to +10% in 2020-2025 (t=1.8, below statistical significance). In 2024, the Tuesday effect actually reversed, producing a t-statistic of -3.95. The anomaly that once captured half the week's returns is being arbitraged away in real time.

Why? The mechanism was likely institutional rebalancing and settlement flows concentrated on Tuesdays. As algorithmic trading and continuous rebalancing spread across the industry, these flows got distributed more evenly across the week. The structural reason for Tuesday's dominance disappeared - not because someone discovered it, but because the market evolved past it.

Daily: S&P 500 Mean-Reversion

The simplest possible daily signal: if the S&P 500 was down today, buy tomorrow. If it was up today, sell tomorrow. Pure lag-1 mean-reversion on ES futures, close to close, 2008 to 2025. No filters, no regime detection, no optimization. One rule.

The full-sample Sharpe is 0.42 with a lag-1 autocorrelation of -0.068. Modest but real. The problem is what happens when you look year by year.

In 2016, this signal produced a Sharpe of 2.65. In 2022, it came back at 1.82. In 2025, another strong year at 1.49. But in between: 2018 was -0.98, 2024 was -0.46, and 2010 was -0.34. The signal alternates between brilliant and worthless with no warning.

This is not decay in the traditional sense. The mechanism - daily mean-reversion driven by overnight position clearing and next-day re-entry - has not disappeared. The lag-1 autocorrelation is still negative on average. But it fluctuates wildly by year, from -0.23 in 2016 (strong reversal) to +0.01 in 2024 (no reversal at all). The signal goes dormant, sometimes for years, then reappears. It is the hardest type of edge to trade: real enough to show up in the data, unstable enough to destroy you if you size for the good years.

The Uncomfortable Truth

The job of a quantitative researcher, at any scale, is not to find a signal. It is to build a process that keeps finding signals after the current ones die. Crabel has been doing this for 40 years. Campbell has been doing it for 50. The signals change. The process compounds.

Crabel said something on Top Traders Unplugged that stuck with me. When asked about the depth of the research problem, whether they had explored enough of the data, he said: "I just feel like we've just scratched the surface. There's so much more to take out of the analysis. If there's one question that's being answered and some development around that, there's going to be another 50 to be asked right after that just on that note alone. It just doesn't end."

Signals die. The work of finding them does not.

Sources and Further Reading

• McLean, R.D. and Pontiff, J. (2016) - "Does Academic Research Destroy Stock Return Predictability?" Journal of Finance, 71(1), 5-32.

• Toby Crabel - Top Traders Unplugged, Episode 152: "Vanishing Edges." Crabel Capital Management founder on 40 years of short-term edge decay.

• Giuseppe Paleologo - Flirting with Models, Season 7, Episode 11: "Multi-Manager Hedge Funds." Former head of enterprise risk at Millennium and head of risk at HRT.

• Kevin Cole - Flirting with Models, Season 5, Episode 12: "Systematic Multi-Strategy with 100+ Models." CEO and CIO of Campbell & Company.

• Russell Korgaonkar - Flirting with Models, Season 4, Episode 15: "Optimizing the Research Process." CIO of Man Group's systematic division on the Red Queen problem.

• Ernest Chan - Flirting with Models, Season 3, Episode 5: "Tail Reaper." On alpha decay in simple models and the democratization of quant trading.

• Rob Carver - Top Traders Unplugged, Systematic Investor Series: Episode 396. Former Man AHL on slow decay in trend following.

• Jeff Rosenberg - Flirting with Models, Season 7, Episode 19: "Systematic Fixed Income." BlackRock on the alpha decay arms race in equities vs fixed income.

• Antti Ilmanen - Flirting with Models, Season 5, Episode 8: "Unexpected Returns." AQR senior researcher on alpha migrating to beta.

• Macrocephalopod - Flirting with Models, Season 6, Episode 5: "Mid-Frequency Crypto Prop Desk." On alpha decay speed at higher frequencies.

• Marcos Lopez de Prado (2018) - "Advances in Financial Machine Learning." Chapter on meta-labeling and signal decay.

• Michael Isichenko (2021) - "Quantitative Portfolio Management." Former Millennium researcher on the practical realities of the pod model.

Related Research

• How I Think About Trading - The decision-making process underneath every backtest and strategy.

• The Delphic Fund - 14 strategies built on structural mechanisms, tracked live.

• Alpha Combinations - Why diversification across uncorrelated signals matters more than any individual Sharpe.

A proprietary trading firm (prop firm) trades its own capital for profit. In the retail prop firm model, you pay a fee to attempt a "challenge": a trial period where you must hit a profit target without breaching loss limits. Pass, and you get a funded account to trade with the firm's money. Fail, and you lose the fee.

Every challenge is the same game: reach a profit target before hitting a loss limit. The rules vary, but the math is universal. This post gives you the tools to evaluate any prop firm challenge, no matter which firm, which rules, which market.

Disclosure: I am not affiliated with any prop firm. No referral links, no sponsorships. This post is purely educational.

The Three Rules Every Challenge Has

Every prop firm challenge boils down to three constraints:

• Profit target (T): How much you must make, typically 6-10% of account.

• Daily loss limit (L): Maximum you can lose in a single day, typically 2-5%.

• Max drawdown (D): Maximum total loss before elimination, typically 4-10%.

Here's how these three numbers vary across the five major firms. The formulas in this post work for any combination.

Step 1: Know Your Numbers

Before any formulas, you need two numbers about your own trading: your average daily return (μ) and your daily volatility (σ). Everything else follows from these.

If you have a track record (50+ trades): calculate your average daily PnL as a percentage of account. That's μ. Calculate the standard deviation of those daily returns. That's σ. Divide μ by σ and multiply by √252 to get your annualized Sharpe. If you did this in Part 1, you already have the number.

If you don't have a track record yet, use the instrument you plan to trade as a guide. Your daily σ is roughly the instrument's daily volatility times your leverage:

Notice how Crude Oil and BTC at 2x leverage are already near the typical 5% daily loss limit. One bad day and you're out. FX traders have the most room.

For μ, be honest. A day trader making 2-3 round trips in NQ with a slight edge might average +0.05% to +0.15% per day on the account. A swing trader holding FX positions for days might average +0.03% to +0.10%. Most people overestimate their μ and underestimate their σ.

Quick calibration: if you're roughly break-even after 3-6 months of trading, your Sharpe is near zero. If you're up 10-15% over a year with occasional 5% drawdowns, you're around Sharpe 0.5-1.0. If you're consistently making money every month with small drawdowns, you might be Sharpe 1.5+.

Does win rate matter? Not directly. A 40% win rate with 3:1 payoff (big wins, small losses) produces the same Sharpe as a 70% win rate with 0.8:1 payoff (small wins, frequent). The formulas below use μ and σ, which absorb both win rate and payoff ratio into a single number. What does matter is how your wins are distributed. Consistency rules (covered below) penalise strategies that make all their money on a few big days, regardless of overall Sharpe.

Step 2: Your Raw Pass Probability

A prop firm challenge is a classic gambler's ruin problem: start at 0, try to reach +T (profit target) before falling to -D (max drawdown). If your daily PnL is:

then the probability of passing is:

For the common case where target equals drawdown (T = D), this simplifies to a sigmoid:

θ is your "edge-to-noise ratio". It captures how good your strategy is (s), how far you must go (T), and how noisy each day is (σ). Here's what the raw numbers look like for a 10% target / 10% drawdown challenge at daily σ = 1.5%:

These raw numbers are too optimistic. The industry average pass rate is ~10%, not 60%. The next section explains why.

Quick sanity check, Calmar ratio: divide the profit target by the drawdown limit. That's the minimum Calmar ratio (annualized return / max drawdown) your strategy needs. To pass reliably, you need roughly double that number. A 10% target with 10% drawdown means minimum Calmar 1.0, realistically ≥ 2.0. A challenge with 8% target and 5% drawdown requires minimum 1.6, realistically ≥ 3.0.

Step 3: What Cuts Your Odds

Five real-world factors reduce the theoretical pass rates. Each one acts independently, and they stack.

Daily loss limits cap your position size

Any single day below the limit terminates you instantly. The limit also constrains your leverage: max leverage = daily limit / (2-3 × instrument vol). For NQ (1.5% daily vol) with a 5% limit, that's about 1.7x, or roughly 3 contracts on $100K. BTC at 3% vol caps you below 1x. You can't bet your way to the target.

Trailing drawdown is much harder than static

Static drawdown fixes the floor at starting balance minus D. Trailing drawdown ratchets the floor up with every new peak. Rally to $115K and the floor is now $105K, not $90K. Pull back to $104K and you're eliminated despite being up. Trailing cuts pass rates by 30-40% versus static.

Consistency rules penalise lumpy winners

Some firms cap your best day at 40-50% of total profit. This means you need 5-7 meaningfully profitable days, not one big hit. A trend-following strategy with Sharpe 1.5 can fail this rule, not because it's unprofitable, but because the rules penalise how it makes money.

Time limits add a lottery element

Some firms impose 30-60 day deadlines. Others have no time limit at all. Subscription-based firms charge monthly, creating implicit time pressure even without a deadline. The impact is large:

A 30-day limit cuts pass rates nearly in half. At Sharpe 1.0, you drop from 70% (unlimited) to 29% (30 days). The expected time to reach a 10% target at Sharpe 1.0 is ~105 trading days. A 30-day window only gives you ~22 trading days. You need to get lucky, not just be good.

Two-phase challenges multiply the penalty

Some firms require two consecutive passes (e.g. Phase 1: 10% target, Phase 2: 5% target, same drawdown).

P(pass both) = P(Phase 1) × P(Phase 2)

Example (Sharpe 1.5, σ = 1.5%): Phase 1 (T=10%, D=10%): ~55%. Phase 2 (T=5%, D=10%): ~80%. Combined: 0.55 × 0.80 = 44%. Phase 2 is easier (lower target with the same buffer) but the multiplication still hurts. Single-phase firms avoid this penalty entirely.

Putting it all together

How do pass rates change across different challenge configurations? The answer depends on the target, drawdown, and your instrument's volatility.

Lower targets are easier (6% vs 10%), and tighter drawdown limits (5% vs 10%) are much harder. The5ers-like setup (T=8%, D=5%) is the hardest challenge at every Sharpe level.

Static vs trailing makes a big difference everywhere, but the gap is largest on tight drawdown limits. On The5ers-like rules (right panel), trailing DD at Sharpe 1.0 drops you from 54% to 38%.

Your instrument matters. FX traders (1% daily vol) have materially higher pass rates than BTC traders (3% daily vol) at the same Sharpe. Lower vol means you're further from the daily loss limit and the drawdown floor on any given day.

After applying all adjustments (daily limits, drawdown type, consistency rules, discrete trading at ~5-10% penalty, and time constraints):

The 10% industry pass rate maps to Sharpe 0.4-0.6. Most challengers have minimal or no edge.

Step 4: Is It Worth the Fee?

The expected value of a single challenge attempt:

EV = P(pass) × P(survive) × Vfunded - Fee

where P(survive) is the probability of surviving to your first payout (~55% within 90 days, based on FTMO's published statistics and aggregated data from trackers such as Prop Firm Match), and Vfunded is what that funded account is worth to you.

Conservative estimate: first payout only

Most funded traders lose their accounts within 90 days. So the realistic baseline is: what's your first payout worth? At Sharpe 1.5, you might make 5% on a $100K funded account over 2-3 months. At 80% profit split, that's $4,000.

Example: Fee = $580, P(pass) = 44% (FTMO two-phase at Sharpe 1.5), P(survive) = 0.55. EV = 0.44 × 0.55 × $4,000 - $580 = +$388. Barely positive. At Sharpe 1.0, this goes negative.

Setting EV = 0 gives the break-even: Sharpe ~1.3. Below that, every fee is negative expected value.

Optimistic estimate: scaling

If you survive the first 90 days, the economics improve dramatically. Most firms offer account scaling:

• The5ers: scales from $100K to $4M with profit split improving from 50% to 100%

• FTMO: scales to $400K with split improving from 80% to 90%

• Apex: 100% split on first $25K of profit, then 90%

A trader who survives 6+ months on a scaling program might extract $20,000-$50,000+ from a single $580 challenge fee. At that point, the EV is overwhelmingly positive, but only ~20% of funded traders last that long. The scaling opportunity is real but shouldn't be the base case for your decision.

The honest comparison: above Sharpe 2.0, the challenge is clearly +EV even on first-payout math. But at that skill level, $50K of your own capital at moderate leverage compounds faster without a profit split, unless you don't have $50K. That's the real use case: prop firms are leverage for undercapitalised skilled traders.

Summary

• Can I pass? Sharpe 0.5: 10-17%. Sharpe 1.5: 40-55%. Sharpe 2.0: 60-75%. Below 0.5, it's a coin flip.

• Trailing or static DD? Trailing cuts your pass rate by roughly a third.

• Two-phase? Phases multiply: 55% × 80% = 44%.

• Min strategy quality? Calmar ≥ 2 × (target / drawdown).

• Max position size? Leverage = daily limit / (2-3 × instrument vol). NQ at 5% limit: ~3 contracts on $100K.

• Consistency rule? Need 5-7 profitable days if best day capped at 40%. Kills trend following.

• Worth the fee? Break-even around Sharpe 1.1-1.5 depending on the firm (1.3 for FTMO, lower for cheaper firms). Below that, every fee is negative EV.

In the next post, I apply all of this to the five major prop firms that survived the 2024-2026 shakeout (FTMO, Topstep, Apex, The5ers, and FundedNext) with firm-specific pass rates, position limits, expected values, a strategy compatibility matrix, and which firms support API access for automated trading.

Recently a friend asked me what to do with their savings. I realised the honest answer requires the same analytical framework I use at work. So I ran the numbers.

The conclusion: the biggest risk isn't a market crash — it's doing nothing. A savings account doesn't just underperform stocks. After inflation, it actually loses money. Here are seven backtests that show why.

The Baseline: Savings Accounts as a Quantitative Proposition

A savings account paying 3-5% per year is not an investment. It is a cash management tool. That is exactly what it should be used for: emergency reserves, short-term liquidity, money you need within 1-2 years.

But many people leave long-term money there. What does that cost?

$10,000 at 3% for 20 years compounds to $18,000. The same $10,000 in a broad equity index has historically compounded to roughly $67,000. That is a $49,000 opportunity cost — and it grows exponentially the longer the horizon.

After inflation, the real return on most savings accounts is close to zero. In many years, it is negative. You are preserving nominal value while losing purchasing power.

Backtest 1: Real Returns After Inflation

Most people think a savings account at least keeps their money safe. In nominal terms, it does — your balance goes up every year. But inflation eats into that balance invisibly.

I ran the numbers using actual historical savings rates (tracking the Fed Funds rate with a lag) and real CPI inflation from 2000 to 2025. Starting with $10,000 at the worst possible moment — right at the dot-com peak:

The savings account went from $10,000 to $8,517 in real terms — you lost purchasing power. The nominal balance grew to $16,500, which feels like a gain, but prices nearly doubled over the same period. The S&P 500 grew to $38,000 in real terms. Even the conservative 60/40 balanced fund reached $22,700. The near-zero interest rate era from 2009 to 2022 was especially destructive for savers — more than a decade of negative real returns.

Backtest 2: Over Any 10-Year Window, Who Wins?

Pick any date since 1993. Hold the S&P 500 for exactly 10 years from that date. What annual return did you get? I did this for every possible starting date — thousands of overlapping 10-year windows — and compared each one to a savings account paying 3%.

Each point on the blue line is the annualised return you would have earned if you bought the S&P 500 on that date and held for 10 years. Green shading means stocks beat savings. Red means savings won. Stocks won 86% of the time — and the only losing windows were those that started right before the dot-com crash and ended during the 2008 financial crisis.

Over 15-year windows, stocks win over 95% of the time. The longer your horizon, the more certain the outcome.

Backtest 3: The World's Worst Market Timer

I built a deliberately terrible strategy: invest $10,000 only at major market peaks — the exact worst moments to buy. Every red triangle below is a purchase at a local all-time high that preceded a 10%+ crash.

$100,000 invested at the 10 worst possible moments since 2007 grew to $290,000. The same money in a savings account: $126,000.

Even the worst market timer in history — someone who only buys right before crashes — still more than doubles their money versus a savings account. The biggest risk isn’t a crash. It’s sitting on the sidelines while compound growth passes you by.

This backtest is inspired by Ben Carlson's "Bob, the World's Worst Market Timer" — one of the most compelling illustrations in personal finance.

Backtest 4: The Drawdown History of the S&P 500

The equity premium exists because markets crash. That is not a flaw — it is the mechanism. If stocks never fell, everyone would own them and the premium would disappear.

The question is not whether you can handle a 10% dip. It is whether you can hold through a 55% drawdown that lasts 13 years to recover — because that has happened within the last 25 years.

Dot-com: −49%. The 2008 financial crisis: −55%. COVID: −34% in five weeks. Rate hikes in 2022: −25%.

Most people sell at the bottom and never re-enter. Behavioural finance research consistently shows that the average investor underperforms the funds they invest in — not because the funds are bad, but because people buy high and sell low.

This is the same problem I face building trading systems: not "what has the highest return in a spreadsheet?" but "what can a real person actually hold through a drawdown without abandoning the strategy?"

Backtest 5: Same Crisis, Five Portfolios

I computed the return and worst-ever drawdown for every common approach to storing long-term money. The data covers 2007 to present — a period that includes the financial crisis, COVID, and the 2022 rate shock.

Equity data is from public market indices. Managed fund returns are from Vanguard, Fidelity, and T. Rowe Price public NAVs. Hedge fund estimates are approximate, based on HFRI Fund Weighted Composite Index public reporting.

The S&P 500 took the biggest hit every time. The All Weather portfolio barely moved in 2008. Bridgewater's Pure Alpha fund actually made 9.4% during the 2008 crisis while the S&P was down 55%. Renaissance's Medallion fund — arguably the most successful hedge fund in history — made money in every crisis: +80% in 2008, +76% during COVID.

The pattern is clear: higher return always comes with bigger drawdowns. There is no free lunch. But the trade-offs vary enormously — and every single option here outperforms a savings account over the long run.

The best hedge funds — Bridgewater Pure Alpha, Renaissance Medallion, D.E. Shaw — deliver returns comparable to or better than the S&P 500 with a fraction of the drawdown. Renaissance Medallion has averaged roughly 39% net annual returns since 1988 with a worst drawdown of about 11%. But these funds charge 2-5% of assets plus 20-44% of profits, and most are closed to new investors. For most people, a low-cost balanced fund delivers the core crash protection at a tiny fraction of the fee.

Backtest 6: A Simple Crash Filter

To illustrate how institutional trend followers manage drawdowns, here is one of the simplest approaches in the industry: the 200-day moving average rule.

Compare the S&P 500 to its 200-day average. Above it: stay invested. Below it: move to cash. This is a backtest, not a recommendation — but the drawdown reduction is striking:

Buy-and-hold worst drop: −55%. With the 200-day rule: −24%. The cost: annual return drops from 10.8% to 9.1%, and you generate roughly 150 round-trip trades over 30 years.

I am not saying this is what you should do. I am showing it because it illustrates a principle that drives most professional money management: you can trade a small amount of return for a large reduction in drawdown.

Balanced funds and All Weather-style portfolios aim at the same outcome through diversification rather than timing. For most people, that is the simpler and more robust approach.

Backtest 7: $500 Per Month Since 2006

Forget the percentages. Here is what actually happens when you invest $500 per month into four different approaches, starting in January 2006 — right before the worst financial crisis in 80 years.

Total invested: roughly $120,000 in each case. The S&P 500 grew to $580K. The 200-day trend rule: $410K. The 60/40 balanced fund: $350K. The All Weather portfolio: $250K.

All four survived 2008, COVID, and every correction in between. The experience of holding through those crashes was completely different. The S&P line shows the highest peak — and the deepest valleys.

Pick the line you could actually live with. That is the only question that matters.

Backtest 8: Is the US Exceptional?

Every backtest above uses the S&P 500. But is that cherry-picking the best market in the world? What if you happened to live in Japan, or Europe, or invested in China?

I ran the same $10,000 investment from 2007 across five major global indices.

The US returned $84,700. Europe and Japan returned $23,000–30,000. China: $25,800. The savings account: $15,000.

The US was exceptional — it returned roughly 3x what other developed markets did. But here is the key: every single market on earth still beat the savings account.

Japan returned 4.2% annually with a 54% drawdown. China returned 4.7% with a 73% drawdown. Even these, the two weakest major markets over this period, turned $10,000 into $23,000–26,000 versus $15,000 in savings.

If you are worried about US exceptionalism fading, the answer is diversification — a global index fund like VT (Vanguard Total World) — not a savings account.

What the Data Says to Avoid

This is not financial advice. The backtests above are historical analyses, not recommendations. Past performance does not predict future results. What follows is what the data from these eight backtests consistently pointed against.

Eight backtests, same conclusion. Here is what each one rules out:

Keeping long-term money in savings. Backtest 1 showed a savings account lost purchasing power over 25 years. $10,000 became $8,500 in real terms. Every other option in every other backtest did better.

Waiting for the right moment to invest. Backtest 2 showed stocks beat savings in 86% of all 10-year windows. Backtest 3 showed that even buying exclusively at market peaks still doubled your money versus cash. The cost of waiting is almost always higher than the cost of bad timing.

Panicking during crashes. Backtest 4 showed drawdowns of 34–55% happen regularly. Backtest 5 showed that balanced portfolios and hedge funds cut those drawdowns in half. Backtest 6 showed a simple trend rule reduces the worst drop from 55% to 24%. The tools exist — the mistake is selling at the bottom.

Ignoring diversification. Backtest 7 showed that dollar-cost averaging into a balanced portfolio survived every crisis since 2006. Backtest 8 showed that the US was exceptional — 3x the return of Japan or Europe — but even the worst major market still beat savings by 2x. A global index fund removes the single-country bet.

Picking individual stocks or using leverage. The SPIVA scorecard shows over 90% of professional fund managers underperform their index over 15 years. If they cannot do it with full-time research teams, a retail investor with a brokerage app has worse odds. Leveraged products amplify losses as much as gains — and human psychology handles losses worse.

Let's find out.

What Your Win Rate Actually Earns You

I computed annual returns for every combination of win rate and payoff ratio (average win divided by average loss), assuming 50 trades per year and 2% account risk on each losing trade. Find your row, find your column:

The number most people think they have — 60% win rate, 1:1 payoff — gives +20%/yr before costs. That's good. But look what happens when losses are slightly bigger than wins (the 0.8x column): it drops to +8%/yr. At 0.5x — where your average loss is double your average win — the same 60% win rate produces -10%/yr.

This is the first surprise. Win rate alone tells you almost nothing. A 45% win rate with 2:1 payoff (+35%/yr) crushes a 65% win rate with 0.8x payoff (+17%/yr). The payoff ratio matters just as much — and most people never measure it.

The small grey text (S=) is the Sharpe ratio — a risk-adjusted measure professionals use. I'll come back to that.

Now Subtract Costs

Those returns are before commissions, spreads, and slippage. Here's what trading actually costs:

• Commissions + spread (~5%/yr): ~0.10% per trade round-trip on stocks. On 50 trades/year that's 5% annual drag.

• Slippage (~3%/yr): you don't always get the price you see. Another ~0.06% per trade on active stocks, more on illiquid names.

• Tax drag (~5%/yr in this example): short-term capital gains can be taxed at income rates. A 20%/yr gross return might become ~15%/yr after tax, depending on your jurisdiction and tax bracket.

The chart starts with the +20%/yr cell from the grid (60% win rate, 1:1 payoff) and strips away each cost. What looked like a solid +20%/yr becomes +7%/yr after all costs - barely beating a savings account. Go back to the grid and subtract ~13% from every cell. Most of the yellow and light green cells turn red.

This is why the bar is higher than it looks. You need to clear a ~13% cost hurdle before any skill shows up in your account balance.

What Actually Happens to Most People

I compiled data from academic studies and broker disclosures to estimate what active retail traders actually earn:

The data sources:

• Barber & Odean (2000): 66,465 US households, 1991-1996. Most active traders underperformed the market by 6.5%/yr. (Journal of Finance, Vol. 55, pp. 773-806)

• Chague, De-Losso & Giovannetti (2020): 1,600 Brazilian day traders tracked for over a year. 97% lost money. Only 1.1% earned more than minimum wage. ("Day Trading for a Living?", SSRN)

• Broker CFD risk disclosures: regulatory disclosures often show that a large share of retail CFD accounts lose money. eToro currently reports 50%; broader broker datasets show materially higher loss rates at many providers.

• DALBAR QAIB 2025 report: In 2024, average equity investor earned 16.5% vs S&P's 25.0% — an 8.5pp gap. Over long windows, average investors have typically lagged broad equity benchmarks by several percentage points per year.

The pattern across all four sources is the same: the majority of active traders lose money outright, and most of the rest underperform the S&P 500 — meaning they'd have been better off buying one index fund and doing nothing.

Why Results Often Differ From Expectations

Win rate is often estimated from memory, not records. Winning trades are easier to remember than small losses, scratched trades, and missed exits. Studies show traders can recall their win rate as 5-15% higher than it actually is. The fix is simple but non-negotiable: track every trade.

Payoff ratio is also easy to underestimate. Many strategies feel good because they win often, but the average loss can still be larger than the average win. That creates a strategy that feels consistent while quietly losing money. Go back to the grid: even 65% win rate at 0.5x payoff loses money.

Volatility makes this harder. A $10K position in a utility stock (0.5% daily vol, moves ~$50/day) and a $10K position in a meme stock (10% daily vol, moves ~$1,000/day) are not the same bet. Without volatility-adjusted sizing, two trades that look equal on screen can have completely different downside.

• S&P 500 ETF (SPY): ~1% daily vol → $10K moves ~$100/day

• Tesla (TSLA): ~3% → same $10K moves ~$300/day

• Bitcoin: ~3-4% → same $10K moves ~$350/day

• Penny stocks: 5-15% → same $10K moves $500-$1,500/day

Stop losses are often placed without reference to volatility. A stop loss exits you automatically when a trade goes against you. Two opposite problems: too tight (1% stop on a 3%/day stock = stopped out by noise, then it rallies without you) and too loose or none (a 5% loss becomes 30% while you wait for recovery). The rule: stop distance ≥ 1.5-2x the stock's daily volatility. A stock moving 2%/day needs a stop at least 3-4% away.

The connection: stop distance determines position size. If your stop is 4% away and you won't risk more than $200 per trade, max position = $200 / 0.04 = $5,000. Volatility sets the stop. The stop sets the size.

The market includes highly resourced professional participants. On the other side of your trade may be a hedge fund with PhD researchers, proprietary data, and millisecond execution. Retail traders are not only trading against other beginners; they are also sharing the market with some of the best-resourced participants in finance.

Fundamentals are only one driver of short-term price. A stock can report great earnings and still drop. Short-term price movements are driven by multiple mechanisms that have nothing to do with whether the company is good or bad:

• Options expiry: on monthly options expiration (third Friday), market makers hedging their gamma exposure create mechanical buying or selling pressure. A stock pinned near a large strike can get pushed there regardless of news.

• End-of-month rebalancing: pension funds and index funds rebalance portfolios at month-end and quarter-end. If equities rallied, they sell stocks and buy bonds to get back to target weights. This creates predictable selling pressure on the last few days of strong months.

• Market maker inventory: when a market maker accumulates too much of one stock, they widen their quotes on the buy side and tighten on the sell side to offload it, pushing the price down even if nothing fundamental changed.

• Forced liquidation: when leveraged traders hit margin calls, their broker sells their positions automatically. This creates cascading selling unrelated to the stock's value - and often at the worst possible time.

• Index rebalancing: when a stock is added to the S&P 500, every index fund must buy it. When one is removed, they must sell. The stock moves 5-10% on the announcement, purely from anticipated flow.

A trader can build a reasonable thesis around fundamentals or a chart pattern and still get stopped out by one of these mechanical forces. The news can be right, the trade can still be wrong, and the difference is often market structure rather than intelligence or effort.

What "Good" Actually Looks Like

The Sharpe ratio (the grey S= numbers in the grid) measures risk-adjusted returns. It's the standard professional benchmark:

Here's what each level looks like as a 1-year equity curve starting with $10,000:

Even Sharpe 1.0 (yellow) has multi-week drawdowns. Sharpe 0.5 (red) — where most retail traders sit — is nearly indistinguishable from the zero-edge grey line. Only at Sharpe 2.0+ does the curve clearly go up.

Go back to the grid. The cells where S ≥ 1.0 are the ones that have a visible upward trend in practice. Below that, you can't tell skill from luck — not for months.

How to Actually Measure Yourself

The single most important thing: track every trade for 50+ trades and calculate two numbers.

1. Win rate: what fraction of trades made money?

2. Payoff ratio: average winning trade / average losing trade.

Look up your cell in the grid. That's your expected annual return — before costs. Subtract ~13% for trading costs (see the waterfall above). If the number is still positive, you might have something real.

If you want the single summary number that professionals use, you can estimate your annualized Sharpe ratio directly from your win rate (p) and payoff ratio (R):

Sharpe = √N × (pR - (1-p)) / (√(p(1-p)) × (R+1))

where N is the number of trades per year. For example: 60% win rate, 1:1 payoff, 50 trades/year = √50 × (0.6×1 - 0.4) / (√(0.6×0.4) × 2) = 7.07 × 0.2 / 0.98 = Sharpe 1.4. That matches the S=1.4 cell in the grid above.

Picker A (Sharpe 0.3) has a positive average but massive variance — the equity curve spikes then crashes. No edge. Picker C (Sharpe 2.7) has tighter, more consistent returns across 50 trades — steady upward curve, +112%.

50 trades is a starting point, not a verdict. With 20 trades, luck dominates. With 50, the signal starts to emerge, but the estimate is still noisy. More trades are better.

The Honest Summary

Most people who think about trading believe they're in the +20%/yr cell (60% win rate, 1:1 payoff). After accounting for the payoff ratio problem, costs, and the gap between perceived and actual win rates, most end up in the -5% to +5% range — underperforming an index fund.

That's not a reason not to try. It's a reason to measure before you commit. Paper trade for 3-6 months, then test at tiny live size because paper trading misses slippage, fills, and psychology. Track every trade. Calculate your win rate and payoff ratio. Look up your cell. If the number after costs is above 10%, you have something worth pursuing. If it's not, the strategy needs work — not more capital.

In the next post, I cover the math of prop firm challenges — how their rules translate into the minimum win rate, payoff ratio, and Sharpe you need to pass.