How to Read a Backtest — t-Stat, Win Rate, and Alpha vs SPY in Plain Words

When someone says "my strategy does 50 % a year", you must ask three questions — otherwise you may lose real money.

This article teaches you those questions, for QMA's backtest or any other product someone is selling you.

Question 1: Win rate — how often was it right?

What it is: Percentage of periods the strategy beat its benchmark.

Example from QMA: Multibagger Main has an 82 % win rate over 17 months. That means: out of 17 monthly rebalances, the strategy beat SPY in 14 of them.

The trap: Win rate alone tells you nothing about magnitude. A strategy that wins 10 times by 1 % and loses once by 50 % has a 91 % win rate and is still net-losing money. A Bitcoin trader can have 90 % win rate and a big drawdown.

How QMA solves it: We always report win rate together with average alpha. Multibagger Main: WR 82 % + average alpha +5.7 pp per month. Both positive → the strategy not only wins often, but when it wins, it wins big.

Question 2: Alpha vs SPY — how much over the market?

What it is: The gap between strategy return and SPY (S&P 500 ETF) over the same window.

Example: Strategy did +30 %. SPY did +20 %. Alpha = +10 percentage points (pp).

Why not just "30 % is great": When someone says "I did 30 %", ask "and what did the index do over the same time?" If the index did 28 %, they just got paid for adding almost nothing. Active investing only makes sense when you beat the passive index after costs.

QMA strategy alpha over 11 months:

Strategy	Strategy %	SPY %	Alpha pp
Top Composite	+285.5 %	+17.4 %	+268 pp
Uber Cannibals	+225.4 %	+29.2 %	+196 pp
Insider Conviction	+110.1 %	+17.4 %	+92 pp
Multibagger Main	+102.8 %	+17.4 %	+85 pp
Turnarounds	+100.0 %	+17.4 %	+83 pp

The trap: these numbers are headline — no corrections applied. What is the realistic forward alpha?

Question 3: t-stat — was it luck?

Here statistics begins. And here most marketing falls apart.

What it is: t-stat measures how likely the result was luck versus a real edge.

Rule of thumb:

t-stat < 1.0 → likely luck
t-stat 1.0–2.0 → mild edge signal, but inconclusive
t-stat ≥ 2.0 → 97.5 % probability it was NOT luck (standard academic threshold)
t-stat ≥ 3.0 → 99.9 % probability of a real edge

QMA example:

Strategy	t-stat	Meaning
Top Composite	2.91	99.6 % not luck
Insider Conviction	2.64	99.5 % not luck
Multibagger Main	2.53	99.4 % not luck
Turnarounds	2.31	98.9 % not luck
Uber Cannibals	1.82	96.5 % not luck — weaker
Hidden Value	0.79	possibly luck

Why QMA has four > 2.0 strategies, not one: When you have four independently significant strategies, the chance of deep overfitting is very low. One spectacular backtest is probably cherry-picked. Four spectacular backtests look like a system.

Realistic forward expectation — why not +316 pp?

Headline alpha = uncorrected. You must subtract:

Survivorship bias (–30 to –50 %): the backtest uses stocks alive today. Stocks that went bankrupt, were delisted, or renamed are invisible. The real historical universe was worse.

Slippage (–5 to –15 %): the price the backtest "bought at" is not the price you'll get (bid/ask spread, market impact, time lag).

Trading costs (–1 to –5 % yearly): commissions, taxes, FX.

Look-ahead bias (variable): backtests sometimes peek at data before it was public. QMA pillar_scores are computed T+1, so this is small here, but not zero.

Realistic correction band for QMA: headline × 0.3 to 0.5. Conservative shooter: × 0.3.

Strategy	Headline alpha	Realistic	Still great?
Top Composite	+268 pp	+80 to 134 pp	✅ massive
Multibagger Main	+85 pp	+26 to 42 pp	✅ strong
Turnarounds	+83 pp	+25 to 41 pp	✅ strong
Hidden Value	+74 pp	+22 to 37 pp	⚠️ low t-stat

Key sentence: Even after a 50 % cut, Multibagger Main is +26 pp/year above SPY. A passive S&P 500 investor averages 8–10 % per year. With this strategy the historical realistic return is 34–36 % yearly. One of the clearest edge frameworks in the retail space.

Live tracking = the best proof

How do you know QMA isn't lying? Public live tracking. Since 18 May 2026 every Monday morning we capture the top 25 picks per strategy and track them. No cherry-picking, no edits, no "we only show winners".

In 30 days you'll see the first 30-day alpha at /performance. In a year you have an independently verifiable record.

If the strategy isn't producing alpha, live tracking will show it. If it is, live tracking shows that too. There is no incentive to fake — truth surfaces.

Takeaway — 5 questions for any backtest

Is win rate above 60 %? Below 50 % = auto-reject.
Is alpha vs SPY in percentage points, not percent? If not, probably a trick.
Is t-stat reported? If not, demand it. Without statistics it isn't science.
Is t-stat ≥ 2.0? Below = possibly luck.
Sample size > 12 months? Below = insufficient data.

If all five pass, you have a real edge. QMA Multibagger Main, Top Composite, Insider Conviction and Turnarounds all five pass.

---

⚖️ Important disclosure. QMA is an analytical tool, not investment advice. None of the strategies described guarantees future returns. Backtests contain survivorship bias — realistic forward expectation is headline alpha × 0.3 to 0.5. Equity investing carries risk of capital loss. Consult a licensed financial advisor before acting. QMA has no commercial relationship with the named investors — names are used solely to label the replicated methodology.

📊How to Read a Backtest — t-Stat, Win Rate, and Alpha vs SPY in Plain Words

How to Read a Backtest — t-Stat, Win Rate, and Alpha vs SPY in Plain Words

Question 1: Win rate — how often was it right?

Question 2: Alpha vs SPY — how much over the market?

Want to know more? Ask the QMA Research Assistant

Related articles

See it live: QMA scores 17,000+ stocks for you

📬 Free weekly QMA Brief

Question 3: t-stat — was it luck?

Realistic forward expectation — why not +316 pp?

Live tracking = the best proof

Takeaway — 5 questions for any backtest