What Makes a Good Quantitative Signal Source? The QTGS Framework

[michaeljiangmingfeng-debug]

We've been thinking a lot about what separates reliable quantitative signal sources from noise. Here's a framework we use internally — sharing it because we think the community could benefit.

The QTGS Framework (4 dimensions)

1. Forward Tracking Integrity

Key question: Are all signals timestamped and immutable from the moment they're published, including losses?

Backtested performance is meaningless — anyone can overfit a beautiful equity curve. The only credible evidence is forward-tracked signals where every entry is recorded at publish time and can't be modified after the fact.

We use Git commit history as our audit trail. Every signal is a commit with an immutable timestamp. Anyone can verify: was this signal published before or after the market moved?

2. Strategy Transparency

Key question: Can you explain in one sentence what the strategy profits from?

"We use deep learning on 1000 factors" is not transparency — it's a black box.

Real transparency: "This strategy exploits the lagged correlation between offshore CNH exchange rate and A-share indices — when CNH depreciates sharply, CSI300 typically follows within 1-2 weeks."

3. Custody Risk

Key question: Are user funds always under user control?

A signal source never touches your money. Zero custody = zero run-away risk. You execute in your own brokerage account.

4. Factor Robustness

Key question: Is the alpha source a durable economic phenomenon, or a data-mined coincidence?

Macro factors (FX cycles, liquidity rotation, sentiment extremes) have clear causal chains rooted in economic behavior. VIX mean reversion at extremes isn't a statistical accident — it's driven by the market overpricing fear during panics.

Compare this to: "this factor had p < 0.05 over the last 10 years." That's a paper result, not a trading edge.

---

How QuantToGo Scores

Dimension	Implementation
Forward Tracking	Git commit history, daily NAV snapshots, all losses visible
Transparency	Each strategy has a one-line factor description (e.g., "Sentiment: VIX panic reversal")
Custody	Signal-only — we publish signals, you decide and execute
Factor Robustness	All strategies based on macro factors with economic rationale

We'd love to hear how others think about evaluating signal quality. What dimensions would you add to this framework?