Combining Indicators: The Multi-Factor Approach

Single Indicators Don't Work

Let's get this out of the way: single-indicator edges are almost non-existent in modern markets.

"Buy when RSI drops below 30" doesn't work. If it did, everyone would do it, and the edge would disappear. Markets are too efficient for simple signals.

But combinations are different. When three independent factors align, you're looking at a confluence that most traders don't see—because most traders aren't combining multiple data sources systematically.

Why Combinations Create Robustness

Single indicators give false signals. A lot of false signals. RSI hits 30, but price keeps dropping. Funding spikes, but the market keeps pumping. One indicator alone lacks context.

Combinations require multiple things to be true simultaneously. Each additional filter reduces false positives. If Factor A has a 60% success rate alone and Factor B has a 60% success rate alone, requiring both might push combined success to 70%+ (if they're measuring different things).

The logic is AND, not OR:

• •Factor A triggered → Maybe trade
• •Factor A AND Factor B triggered → Probably trade
• •Factor A AND Factor B AND Factor C triggered → Strong setup

Each additional confirmation improves signal quality at the cost of signal frequency.

Confirmation vs Redundancy

Here's a critical distinction: you want confirmation, not redundancy.

Redundancy: Combining RSI with Stochastic RSI. They measure the same thing (momentum oscillators). If one is oversold, the other probably is too. No new information.

Confirmation: Combining funding rate with open interest with liquidation levels. Each measures something different: sentiment, positioning, and stop placement. New information from each.

Redundant factors don't improve signals—they just reduce frequency without increasing quality. Confirmatory factors each add independent information.

When selecting indicators to combine, ask: "Does this measure something the other indicators don't?" If yes, potential combination. If no, redundancy.

The Optimal Number of Factors

How many factors should you combine? Based on our testing:

1 Factor: Almost never works in isolation. Too many false signals.

2 Factors: The minimum viable combination. Primary condition + confirmation. Often sufficient.

3-4 Factors: Sweet spot for most edges. Strong confirmation without over-fitting.

5+ Factors: Diminishing returns. Each additional factor dramatically reduces signal frequency. Risk of curve-fitting to specific historical patterns.

We typically target 2-3 factor combinations. Enough filters to improve quality, not so many that signals never fire.

AND Logic vs OR Logic

Most robust edges use AND logic: all conditions must be true.

AND example: "LONG when funding_z < -1.5 AND oi_rising AND price_at_liquidation_level"

OR logic can work but requires careful construction:

OR example: "LONG when (funding_z < -2.0 OR cvd_z < -2.5) AND regime = bull"

OR within a category (any of these extreme readings) combined with AND across categories (but must be in this regime). Hybrid approaches can capture more setups while maintaining quality.

Be careful with pure OR logic—it degrades into "any excuse to trade" quickly.

Avoiding Correlation Between Factors

Factors can be mathematically correlated even if they seem conceptually different.

Example: Funding rate and long/short ratio often move together. High funding usually means more longs. Combining them feels like two factors but might only be one and a half in terms of independent information.

During discovery, we check correlation matrices between indicator z-scores. If two indicators correlate > 0.7, combining them adds limited value.

We prefer factors from different "families":

• •Derivatives (funding, OI, liquidation levels)
• •Flow (CVD, volume delta, taker buy/sell)
• •Structure (price levels, support/resistance, trends)
• •Sentiment (long/short ratio, whale positioning)

Cross-family combinations tend to be less correlated and more robust.

Primary + Secondary Structure

A useful framework: designate one factor as primary and others as secondary confirmations.

Primary Factor: The main signal generator. Defines "something is happening."

Secondary Factor(s): Confirmation filters. Defines "yes, this is likely real."

Example:

• •Primary: Funding z-score drops below -2.0 (extreme oversold sentiment)
• •Secondary 1: Open interest rising (new positions being opened, not just existing longs closing)
• •Secondary 2: Price at major liquidation cluster (mechanical buy pressure nearby)

The primary factor generates the alert. Secondary factors confirm whether to act.

This structure helps you think about what you're actually trading. The primary factor is your thesis; secondary factors are your evidence.

Grid Search for Combinations

With 50+ indicators, manually testing combinations is impossible. You need systematic search.

Grid search approach:

1. •Define threshold levels for each indicator (e.g., z-score of -2, -1.5, -1)
2. •For each primary indicator × threshold combination
3. •For each secondary indicator × threshold combination
4. •Test the combined edge
5. •Record results in database

This creates millions of candidates, which is exactly what you want. The discovery pipeline filters most out—survivors are genuinely robust.

For efficiency, we use hierarchical search:

1. •Test all single-indicator edges first
2. •Only combine indicators that show some solo potential
3. •Skip combinations where neither indicator works alone

This reduces search space dramatically while preserving likely winners.

Example: Building a Multi-Factor Edge

Let's walk through building an actual edge:

Hypothesis: In bull markets, extreme negative funding combined with OI increase and CVD divergence indicates capitulation selling before a bounce.

Primary Factor: Funding z-score < -2.0

Secondary Factors:

• •OI change z-score > 0.5 (positions being opened)
• •CVD z-score < -1.5 (selling pressure extreme)

Regime Filter: Price above 200-day MA (bull market only)

Test Results (hypothetical):

• •Funding alone: 54% win rate, PF 1.1
• •Funding + OI filter: 61% win rate, PF 1.6
• •Funding + OI + CVD filter: 68% win rate, PF 2.3
• •With regime filter: 73% win rate, PF 3.1

Each layer improved quality. Final combination is tradeable; components alone were not.

The Combination Documentation Standard

Every promoted edge should document:

1. •Primary Factor: What and why
2. •Secondary Factors: What and why for each
3. •Logic: How factors combine (AND/OR)
4. •Regime: Required market conditions
5. •Thresholds: Exact z-score levels for each factor
6. •Rationale: Market microstructure explanation
7. •Backtest Stats: By regime and overall
8. •Correlation Check: Confirmation that factors aren't redundant

This documentation prevents "mystery edges" where you don't remember why something was supposed to work.

Key Takeaways

1. •Single indicators rarely create tradeable edges—combinations are necessary
2. •Combine for confirmation, not redundancy—each factor should measure something different
3. •2-4 factors is typically optimal—enough filters without over-fitting
4. •Use AND logic as default, OR logic carefully
5. •Check correlation between factors—highly correlated factors add less value
6. •Structure as primary (signal) + secondary (confirmation)
7. •Grid search systematically—too many combinations for manual testing
8. •Document the rationale for every combination

The art of edge discovery is largely the art of finding indicator combinations that capture real market dynamics. Get this right, and you have signals that work. Get it wrong, and you have noise masquerading as strategy.