Alpha signal research
From signal to portfolio
Systematic alpha research on US equities and China A-shares. Signals are discovered and validated on a self-built platform, then combined into market-neutral portfolios. Signals and implementation stay private; results are published per portfolio.
In developmentResearch status →
Three questions worth asking
Research status: this is a work in progress
Training, validation and test periods, month by month
The three periods are split by time. Green months are gains and red months are losses; darker means larger.
Monthly return, %
Rolling 12-month Sharpe
Annualised Sharpe over the past 252 trading days. A flat, positive line means performance is not concentrated in one episode.
What happens with more capital
The same signals, re-run with market impact for each fund size.
From thousands of candidate signals to a small library
Very few candidates survive the tests. Signals appear here only as anonymous aggregates.
Pairwise correlation between library signals
Library growth, portfolio Sharpe after costs
In research
One platform covers the whole research loop
Every stage, from data to publishing, is written in-house and shared across markets. Research, testing, backtesting and publishing all run automatically.
Research principles
- No look-ahead. Every signal and portfolio is checked automatically so a backtest never uses information that wasn't known at the time.
- No survivorship bias. On each historical day a portfolio only picks from stocks that actually existed that day.
- Out-of-sample testing. Results are split by time, and the most recent stretch is held out from all research for a final test.
- Control for data mining. The more candidates a round tests, the higher the bar; redundant or decayed signals are retired.
Limits
- Everything here is a historical backtest, not live trading. The cost model is detailed but still an estimate of real execution.
- The test period was seen during research, so it is a second check; the true out-of-sample test is the held-out period.