Quantitative thinking. Simplified.

Data-driven observations, charts and quantitative analysis.

Featured insight

5 Sep 2026

From data to signal: building robust models.

An overview of a quantitative research process: data, feature engineering, modelling and validation.

Read the full insightIllustrative article

Market data

  • Price
  • Volume
  • Open interest
  • Technical indicators
  • Derived features

Feature engineering

Changes, not levels · scaled for the regime · no future information

Model

Regularised regression · sequence models

Signal

Judged out of sample, rolling forward
Conceptual diagram of a research process. Model parameters and thresholds are not published.

Data & statistics

Fat tails

Simulated: 3,000 daily returns with fat tails, scaled to 1% daily volatility. Not market data.

  • Simulated returns
  • Normal, same volatility
0%10%20%30%−4%−2%0+2%+4%Daily returnBEYOND ±3%: 1.4% OF DAYSNORMAL DISTRIBUTION: 0.27%
View data table
Share of simulated days by daily return
Daily returnSimulatedNormal
−4.0% to −3.5%0.3%0.0%
−3.5% to −3.0%0.4%0.1%
−3.0% to −2.5%0.6%0.5%
−2.5% to −2.0%1.2%1.7%
−2.0% to −1.5%2.7%4.4%
−1.5% to −1.0%7.0%9.2%
−1.0% to −0.5%14.9%15.0%
−0.5% to 0.0%24.1%19.1%
0.0% to +0.5%23.1%19.1%
+0.5% to +1.0%14.0%15.0%
+1.0% to +1.5%6.6%9.2%
+1.5% to +2.0%2.3%4.4%
+2.0% to +2.5%1.7%1.7%
+2.5% to +3.0%0.5%0.5%
+3.0% to +3.5%0.2%0.1%
+3.5% to +4.0%0.2%0.0%

Diversification and correlation

Volatility of an equal-weight portfolio, as a share of one strategy's, for three average correlations. Computed from the formula, not results.

  • Correlation 0
  • Correlation 0.2
  • Correlation 0.5
0%25%50%75%100%12345678910Number of strategies4 UNCORRELATED: HALF THE VOLATILITY32% · ρ 053% · ρ 0.274% · ρ 0.5
View data table
Portfolio volatility as a share of a single strategy's
StrategiesCorrelation 0Correlation 0.2Correlation 0.5
1100%100%100%
271%77%87%
358%68%82%
450%63%79%
545%60%77%
641%58%76%
738%56%76%
835%55%75%
933%54%75%
1032%53%74%

What a real signal looks like

Simulated: 260 observations of a signal with a true correlation of 0.12 to the next-period return. Not market data.

  • Observations
  • Fitted line
-3-3-2-2-1-100112233Signal, standard deviationsNext-period returnCORRELATION 0.13 · EXPLAINS 1.8%
View data table
Summary of the simulated signal
MeasureValue
Observations260
True correlation0.12
Sample correlation0.13
Variation explained1.8%
Fitted slope0.14

Models & methods

Quantitative library

View all

Mathematics. Data. Markets.

A systematic approach to opportunity.

We use quantitative methods to find structure in market data, and build strategies that can be tested and repeated.

See the published record