PERFORMANCE · HONESTY
CAGR, Sharpe, Drawdown: how to honestly read a backtest
~7 min read · published on May 30, 2026 · critical reading
"Our portfolio returned +30% per year over the last 5 years."
An apparently impressive sentence. In practice, almost unusable. Without five more data points, it's impossible to know if this is (a) genuinely good management, (b) luck, (c) statistical illusion, or (d) pure overfitting. This post shows the numbers that matter and — perhaps more relevantly — what you should ask before believing any of them.
The numbers that matter
CAGR — Compound Annual Growth Rate
The compound annual growth rate. It is the return equivalent per year that, when compounded, takes you from initial to final value. Unlike the arithmetic average of annual returns — which overestimates when there is volatility — CAGR is what you actually achieve.
Caution 1: CAGR depends critically on the start and end points. If you start the backtest in February/2009 (crash bottom) and end in January/2021 (pre-high Selic peak), the CAGR of any stock portfolio will look magical. Always ask: what is the full period? Does it include at least one bear cycle?
Caution 2: CAGR alone says nothing about risk. A portfolio with CAGR +20% and maximum drawdown of -70% is very different from one with CAGR +14% and maximum drawdown of -25% — the second is objectively better for anyone who needs the money in some window.
Sharpe ratio
Excess return over the risk-free rate, divided by volatility. In Brazil, risk-free rate = CDI. (CAGR − CDI) / annualized volatility.
Sharpe above 1.0 is rare and very good. Above 1.5 suggests something between "exceptional strategy" and "backtest with bias". IBOV historically runs Sharpe between 0.3 and 0.6 — so anything well above 1.0 over a long window deserves technical skepticism before admiration.
VORTEX QSP delivers a Sharpe of 0.96 in 7.3 years walk-forward. Significantly above IBOV (~0.53), but within plausible range for multifactor strategy — it's not the kind of number that screams "overfitting".
Sortino ratio
Variant of Sharpe that penalizes only downside volatility (downside deviation), not upside. Makes more economic sense — investors don't complain about positive volatility. Sortino above Sharpe is typical of asymmetrically favorable strategies.
Max drawdown
The largest peak-to-trough drop of the capital curve during the entire period. It's the number that hurts most emotionally when it happens. Strategies with smaller drawdowns have much higher adherence rates — because the investor doesn't give up midway.
VORTEX QSP had Max DD of -33.2% in 2020-2022 (pandemic + interest rate shock). IBOV had -46.8% in the same period. 14 percentage points less is the difference between "portfolio that hurts but you can tolerate" and "portfolio you liquidate at the bottom".
Information ratio
Alpha over the benchmark, divided by tracking error. Measures consistency of outperformance. IR > 0.5 is already considered excellent in equity quant; > 1.0 is extremely rare.
Beat rate
Percentage of windows (years, months) in which the strategy beat the benchmark. VORTEX QSP beats IBOV in 6 of 8 years — beat rate of 75%. Beat rates of 100% should raise suspicion; the market has enough noise that nobody wins all the time.
The three questions that separate honesty from fiction
1. What is the universe?
A strategy "that returned 25% per year buying Brazilian small caps" is radically different from one that returned the same buying IBOV blue chips. Small caps have more potential alpha but also less liquidity, more slippage, and the backtest may be exaggerated because stocks that went bankrupt were retroactively removed from the universe.
VORTEX QSP uses IBrX-100 universe with minimum liquidity. It's a conservative universe — alpha may be lower than in small caps, but it's replicable in real life without relevant market impact.
2. What is the period?
Too short a period (3-5 years) doesn't cover different regimes. Too long (10+) risks hindsight bias. Empirical sweet spot: 7-10 years covering at least one bull and one bear. VORTEX QSP covers 7.3 years including pandemic, interest rate shock, two administrations.
3. What is the exit rule?
Every strategy fails. The question is: when the sequence of failures arrives, what happens? Strategies with disciplined rebalance logic (monthly, fixed rules) tend to perform as the backtest suggests. Strategies that depend on "manager's feeling" to enter and exit rarely perform what they say.
Red flags
Beware especially of:
- Backtest without costs. Serious industry discounts 10-20 bps per side in liquid equities. Without this, any number is inflated by 1-3 percentage points per year.
- "Strategy tested over 30 years with return X%". If the strategy was formulated in 2023 and tested over 30 years until 2023, it's overfitting by construction. The only honest test is forward — after the formulation date.
- Charts without scale. Capital curve without proper Y-axis is visual manipulation. Always demand month-by-month breakdown.
- Focus on absolute return without comparing to IBOV+CDI. +15% p.a. is different in a 13% Selic environment (CDI wins on its own) and 2% Selic (CDI falls short).
- No period of underperformance. Every active strategy has windows where it loses to the benchmark. If the backtest shows perfect outperformance every month, assume cherry-picking.
How VORTEX QSP presents itself
The Performance page publishes all numbers — CAGR, Sharpe, Sortino, Max DD, beat rate, information ratio — with complete month-by-month breakdown, unfiltered. Good months and bad months appear the same way. Including the two years that VORTEX QSP lost to IBOV.
Walk-forward without look-ahead is described in detail in the post Walk-forward backtest: how not to fool yourself.
To close
Historical performance does not guarantee future performance. This disclaimer is true but insufficient. The most important question is not whether the past repeats — it's whether the method is replicable, auditable, and disciplined. Rigorous backtests (walk-forward, frozen parameters, realistic costs) are a necessary condition. They don't give guarantees, but they separate serious conversation from marketing.
Everything VORTEX QSP publishes passed these criteria. It doesn't mean it will work in the next cycle. It means that if it doesn't, it will be because the market changed structurally — not because the backtest was an illusion.
Explore the products
Discover VORTEX QSP → https://www.vortexqsp.com.br
Top comments (0)