Quantum Fund 2 · Quantitative Research Lab
Three years of work replicating what a full floor of quant traders used to do — now run every day by our own research engines.
0 years
of development
~USD 0
invested only in building the backtest engines
0 years
of daily market data per test
0
independent strategy trials in the current cohort
0
hypotheses in the research queue
0
engineering releases
01 · What QF2 does
Every day QF2 searches the internet for the best hypotheses and data it can find — peer-reviewed papers, academic working papers, practitioner research, alternative datasets — and turns them into institutional-level backtests.
Building these engines cost close to USD 40,000, replicating the work a whole floor of quant traders did a few years ago.
01
Discover hypothesis
02
Acquire & clean data
03
Backtest scientifically
04
Promote or kill
↻
02 · The scientific method
State a falsifiable hypothesis before touching data.
In-sample / out-of-sample split, with an untouched hold-out period.
t-test on returns, t-statistic and p-value; we require t > 3 (Harvey, Liu & Zhu multiple-testing hurdle).
CPCV across many paths.
Monte Carlo resampling of trades.
Tested against data snooping.
Penalizes the number of trials we ran.
If Sharpe decays >50% or the in-sample/out-of-sample CAGR ratio is >2x, the strategy is flagged as data-mined and killed.
Real example
A strategy showed an in-sample Sharpe of 1.88. Out of sample it fell to 0.49. Verdict: likely data-mined — rejected. Most ideas die here. That is the point.
03 · Metrics we measure
01
Compound annual growth rate of capital.
02
The deepest peak-to-trough decline.
03
Excess return per unit of volatility.
04
Like Sharpe, but penalizes only downside volatility.
05
CAGR divided by maximum drawdown.
06
Sharpe corrected for skew and kurtosis.
07
Probability the true Sharpe exceeds a benchmark.
08
PSR penalized for the number of trials run.
09
How unlikely the result is under pure chance.
10
Size of the right tail versus the left tail.
11
How much results depend on a few bets.
12
Gross profits divided by gross losses.
13
How often we win and the average expected gain per trade.
14
Van Tharp's system quality, in units of risk.
15
How much performance degrades on unseen data.
16
Stability of results across validation paths.
04 · Top 10 systems
Ranked by our composite robustness score. Hypothetical backtests on daily data, unlevered, each system stand-alone (before portfolio combination).
CAGR vs. max drawdown
Deflated Sharpe already penalizes for the 114 independent trials we ran. Drawdowns shown are stand-alone; combining uncorrelated systems is the next layer of the process.
05 · Advisory
QF2 is advised by the team behind an established program for quantitative trading with Python — specialists in systematic strategies, financial data science and production backtesting. Through recurring strategy clinics and system reviews, they challenge our hypotheses, our code and our validation before anything is promoted.
01
Recurring sessions to stress-test new hypotheses and research ideas.
02
Independent review of engines, data pipelines and validation methods.
03
Institutional standards for risk, robustness and deployment.
Advisor names withheld at their request and to protect proprietary research.
06 · Founder
Founder · Quito, Ecuador