Most traders believe the edge comes from indicators.

Professional quantitative traders know the edge comes from features.

A moving average, RSI, MACD, or any other indicator is simply a mathematical transformation of price. If thousands of traders are using the same transformation, there is little reason to expect it to consistently produce an advantage.

Feature engineering is different.

Feature engineering is the process of creating measurable variables that describe what the market is actually doing. Instead of asking, "Is RSI oversold?", a quantitative trader asks, "What characteristics consistently existed before profitable trades?"

That shift in thinking changes everything.

What Is a Feature?

A feature is simply a measurable piece of market information that a trading system can analyze.

Examples include:

• Distance from session VWAP

• Order block age

• Number of liquidity sweeps during the last hour

• Relative volume

• Volume imbalance

• Time since the last swing high

• Fair Value Gap size

• Opening Range width

• ATR expansion

• Delta divergence

• Trend persistence

• Consecutive bullish candles

• Tick velocity

• Volume Z-score

Each feature describes one aspect of market behavior.

One feature alone rarely creates an edge.

The interaction between dozens of features often does.

Markets Are Multi-Dimensional

Price is only one dimension.

Professional systems analyze many dimensions simultaneously.

Examples include:

Time Features

• Minute of day

• Day of week

• Session

• Time since market open

• Time since last volatility expansion

Price Features

• Distance from previous day's high

• Distance from overnight high

• Distance from VWAP

• Position inside daily range

• Trend slope

• Pullback depth

Volatility Features

• ATR

• Range compression

• Volatility expansion

• Standard deviation

• Realized volatility

• Intrabar volatility

Volume Features

• Relative volume

• Participation rate

• Volume acceleration

• Volume spike

• Delta imbalance

• Cumulative delta

Structure Features

• Liquidity sweep

• Fair Value Gap

• Market Structure Shift

• Break of Structure

• Order Block interaction

• Range breakout

• Rejection wick percentage

Behavioral Features

These describe how the market behaves rather than where it is.

Examples:

• Number of failed breakouts

• Consecutive closes in one direction

• Momentum decay

• Trend exhaustion

• Speed of reversal

• Expansion after compression

These often contain significantly more predictive information than traditional indicators.

Why Feature Engineering Works

Markets are generated by human behavior.

Human behavior leaves patterns.

Patterns create measurable characteristics.

Those characteristics become features.

Your job is not to predict price.

Your job is to discover which measurable characteristics consistently appear before profitable outcomes.

Good Features vs Bad Features

Weak Feature

RSI < 30

This is a fixed mathematical threshold used by millions of traders.

Better Feature

Price has fallen more than 1.8 ATR while volume participation remains below the 20-day median.

Now you're describing market behavior instead of an indicator.

Even Better

Price has:

• Swept previous day's low

• Filled 80% of an unfilled Fair Value Gap

• Relative volume > 2.5

• Delta shifts positive

• First higher low forms within three candles

This combination describes an actual market event.

The Power of Feature Combinations

No single feature may have an edge.

But combinations often do.

Example:

Feature A

• Relative Volume > 2

Win Rate: 54%

Feature B

• Liquidity Sweep

Win Rate: 53%

Feature C

• ATR Expansion

Win Rate: 52%

Individually these are mediocre.

Together:

A + B + C

Win Rate: 71%

This is where quantitative trading begins.

Categories of Powerful Features

1. Event Features

Did something happen?

Examples:

• BOS occurred

• Liquidity sweep occurred

• Volume shock occurred

• Gap filled

• Opening Range broke

2. State Features

What state is the market currently in?

Examples:

• Trending

• Balanced

• Expanding

• Compressing

• Mean reverting

3. Context Features

Where is the event happening?

Examples:

• Premium

• Discount

• Previous day's range

• Weekly high

• Session high

• Overnight low

4. Interaction Features

These combine variables.

Examples:

High Relative Volume AND

Large Fair Value Gap

Small Fair Value Gap AND

Volatility Compression

Liquidity Sweep AND

Low Participation

Interaction features frequently outperform the individual variables.

Creating Features

Professional researchers rarely start with indicators.

Instead they ask questions.

Examples:

What happens after:

• unusually large candles?

• three consecutive failed breakouts?

• five minutes of declining volume?

• expanding volatility?

• aggressive buying into resistance?

• compressed ranges?

Every question can become a feature.

Transform Raw Data

Raw market data includes:

• Open

• High

• Low

• Close

• Volume

Everything else is engineered.

Examples:

Body %

Upper wick %

Lower wick %

Range percentile

Volume percentile

Delta ratio

ATR multiple

Gap percentage

Distance to liquidity

Swing age

Impulse strength

Momentum persistence

These transformations often contain more useful information than the raw inputs.

Feature Importance

Once hundreds of features exist, determine which ones actually matter.

Methods include:

• Information Gain

• Mutual Information

• SHAP values

• Permutation Importance

• Random Forest Importance

• Logistic Regression coefficients

Many popular trading ideas contribute almost nothing.

Unexpected features frequently become the strongest predictors.

Remove Redundant Features

If two features measure nearly the same thing, keeping both adds little value.

Examples:

• EMA20

• EMA21

These are almost identical.

Instead prefer independent information:

• Relative Volume

• Trend slope

• Session timing

• Volatility regime

Independent features improve model quality.

Avoid Look-Ahead Bias

Every feature must be available at the exact moment the trade is entered.

Examples of invalid features:

• Maximum Favorable Excursion

• Final candle close

• Trade outcome

• Future volatility

• End-of-day range

If a feature uses future information, the strategy is not tradable.

Continuous Feature Discovery

Feature engineering never truly ends.

Professional firms continuously generate thousands of new candidate features.

Typical workflow:

1. Generate a hypothesis.

2. Engineer the feature.

3. Validate it on historical data.

4. Test robustness across different markets.

5. Remove unstable features.

6. Combine surviving features into a model.

7. Validate with out-of-sample, walk-forward, Monte Carlo, and live forward testing.

8. Repeat.

The edge compounds over time as the feature library grows.

The Quant Mindset

Retail traders search for the perfect indicator.

Quantitative traders build better descriptions of the market.

The market is not driven by indicators—it is driven by order flow, liquidity, volatility, participation, timing, and trader behavior.

Feature engineering transforms those forces into measurable variables.

The greater your ability to describe the market with meaningful features, the greater your ability to discover statistical edges that others overlook.

In modern algorithmic trading, the quality of your features often matters far more than the sophistication of your model.

The model learns from the information you provide. If the features capture genuine market behavior, even simple models can produce remarkable results. If the features are weak, no amount of optimization or machine learning will create a lasting edge.

Ultimately, successful trading systems are not built on better indicators—they are built on better questions, better measurements, and better features.

Learn More In Stat Discord.

Sign up for “Quant Edge” Below to Join ⬇️