SimicX Quant Research Framework

How we evaluate whether an alpha factor actually works

Every alpha published through SimicX has passed this exact 9-stage pipeline. Our approach is built on what is actually defensible in practice — rank-based testing, cost-aware validation, sector-neutral checks, and out-of-sample discipline.

Walk-forward validation

Expanding windows with quarterly retraining — no single lucky period drives conclusions.

Cost-adjusted metrics

Per-trade commission, slippage, and spread applied at every rebalance.

Sector-neutral testing

Z-score within GICS sectors to confirm genuine within-sector stock selection.

Out-of-sample discipline

No alpha ships on in-sample beauty alone.

9-Stage Pipeline

Collect raw scores

One lagged score per stock per date — no leakage, no survivorship bias.

Rank cross-sectionally

Stable ordinal preference replaces fragile raw magnitudes.

Compute IC

Multi-horizon Spearman IC at 1, 5, 10, 20, and 40 days — full decay profile.

Test significance

Newey-West adjusted t-stats account for autocorrelation in IC series.

Check quintiles

Monotonic bucket returns confirm real economic structure.

Build simple L/S

Gross long-short portfolio before any optimization complexity.

Neutralize

Z-score within GICS sectors to verify pure stock-selection signal.

Apply costs

Per-trade commission, slippage, and spread applied at execution time.

Validate OOS

Walk-forward on unseen data — no alpha passes without this gate.

Main Metric

Spearman IC

Ranks are less sensitive to outliers and better reflect real portfolio ordering.

Core Reality Check

Costs + OOS

No factor survives if it cannot pass per-trade execution costs and out-of-sample tests.

Strict Framing

Heuristics, not promises

Research discipline for triaging factors — not guarantees of live performance.

Examples use simulated yet realistic values — logic is production-accurate.Newey-West lag is our default; alternative bandwidths exist.Transaction costs are applied per-trade (commission + slippage + spread) at each rebalance.

Overview

What we measure, and what we refuse to overstate

At SimicX, a factor is just a score per stock per date. The score is not yet a trade, not yet a portfolio, and not yet a claim about capacity. Our research keeps those layers strictly separate.

Factor score

One scalar per stock per rebalance date. It can come from fundamentals, technicals, alternative data, or a model — our pipeline handles all types.

Rank first, magnitude second

We default to rank IC because the ordering matters more than the raw spacing. This is consistent across all SimicX alpha reports.

Portfolio is a later layer

Long-short returns, constraints, turnover, and risk model choices belong after signal evaluation. We never conflate the two.

Thresholds are heuristics

A mean IC of 0.05 can be useful. A mean IC of 0.20 can be brilliant or suspicious. Our team applies context, not blind rules.

Running Example

We follow one simulated factor called 12M Earnings Revision on an illustrative 500-stock US large-cap universe. The six names below are only a teaching slice of that wider universe.

Stock	Sector	Raw score	Cross-sec rank	Bucket	Fwd return
AAPL	Tech	+1.42	489 / 500	Q5	+5.1%
MSFT	Tech	+0.87	398 / 500	Q4	+3.4%
GOOGL	Comm.	+0.21	287 / 500	Q3	+2.1%
AMZN	Cons. Disc.	-0.33	198 / 500	Q2	+0.6%
META	Comm.	-1.05	47 / 500	Q1	-1.8%
TSLA	Cons. Disc.	-1.88	12 / 500	Q1	-4.2%

The values above are simulated to illustrate our workflow honestly. Every alpha we publish goes through this same structure with real market data.

Explore live alphas on Alpha Stream

Navigate