Understanding Correlation Strength (R-Values)

Educational Resource: This guide explains statistical correlation analysis. These metrics describe historical relationships and do not predict future outcomes. All investments involve risk and past performance does not guarantee future results.

What is a Correlation?

A correlation measures how strongly two things move together in historical data. In trading, we use it to measure relationships like:

Insider buying and future price increases
8-K filings and volatility changes
Momentum signals and continued trends

The correlation coefficient (called "r" or "r-value") tells you how reliable this relationship was historically.

Think of it like measuring how well two dancers move in sync: r = 1.0 means perfect synchronization, r = 0 means no coordination at all.

The R-Value Scale

Correlation coefficients range from -1.0 to +1.0:

Positive Correlations

r = +1.0: Perfect positive relationship

When X increases, Y always increases by the same proportion
Essentially never happens in real markets

r = +0.7 to +0.9: Very strong positive

When X increases, Y tends to increase
Highly reliable relationship
What we look for in momentum patterns

r = +0.5 to +0.7: Strong positive

Clear upward relationship
Good quality signal
Acceptable for most strategies

r = +0.3 to +0.5: Moderate positive

Noticeable relationship
Requires additional confirmation
Minimum threshold for consideration

r = 0 to +0.3: Weak positive

Barely detectable relationship
Not actionable
We filter these out

Negative Correlations

Negative correlations are just as useful - they indicate inverse relationships.

r = -0.7 to -1.0: Very strong negative

When X increases, Y decreases
Excellent for mean reversion strategies
"Fade the move" setups

r = -0.5 to -0.7: Strong negative

Clear inverse relationship
Good for contrarian plays

r = -0.3 to -0.5: Moderate negative

Noticeable inverse pattern
Use with additional signals

Example of negative correlation:

    
    overbought_RSI → forward_return_5d (r=-0.62)

Translation: When RSI signals "overbought," the stock tends to decline over the next 5 days. This is a fade signal, not a chase signal.

Interpreting R-Values in Drivers

Let's decode real examples you'll see:

Example 1: Strong Positive Correlation

    
    insider_buying_20d → forward_return_30d
r=0.74, n=156

What r=0.74 means:

Very strong positive relationship
When insiders bought, stocks rose ~74% of the time historically
High-quality pattern worth your attention

How to use it: Look for opportunities where this driver contributes ≥20% to the total score. Combined with high sample size (n=156), this is a reliable signal.

Example 2: Moderate Positive Correlation

    
    earnings_momentum → forward_return_20d
r=0.42, n=203

What r=0.42 means:

Moderate positive relationship
Pattern exists but not overwhelming
Needs confirmation from other drivers

How to use it: Don't trade this pattern alone. Look for 2-3 other drivers with similar direction to increase confidence.

Example 3: Strong Negative Correlation

    
    high_volatility_spike → forward_return_10d
r=-0.58, n=134

What r=-0.58 means:

Strong inverse relationship
After volatility spikes, stocks historically calmed down
Mean reversion pattern

How to use it: This is a contrarian signal. When volatility explodes, the pattern suggests it won't last. Consider selling volatility (options strategies) or waiting for entry.

Example 4: Weak Correlation

    
    volume_increase → forward_return_30d
r=0.21, n=287

What r=0.21 means:

Weak relationship
Volume alone doesn't predict much
Not actionable as standalone signal

How to use it: You won't see this in our system - we filter correlations below r=0.3. If you see it in external research, ignore it unless combined with stronger signals.

What Makes a Correlation Strong?

R-value alone doesn't tell the whole story. Evaluate three factors together:

1. Absolute Strength (|r|)

Use absolute value to compare positive and negative correlations:

|r| = 0.70 is equally strong whether r = +0.70 or r = -0.70
Direction matters for strategy, but strength is what matters for reliability

2. Sample Size (n)

A correlation of r=0.80 based on n=15 observations is less reliable than r=0.60 based on n=200 observations.

Why? Small samples are prone to luck. Large samples prove the pattern is real.

Learn more: Sample Size Importance

3. Statistical Significance (p-value)

The p-value tells you the probability the correlation happened by random chance.

p < 0.05: Less than 5% chance it's random (we require this)
p < 0.01: Less than 1% chance it's random (very strong)
p < 0.001: Less than 0.1% chance it's random (extremely strong)

Learn more: Statistical Significance

The Sweet Spot: R-Values We Look For

Our quality standards for showing you patterns:

Minimum Requirements

|r| ≥ 0.3: Moderate correlation or better
n ≥ 30: Minimum sample size
p < 0.05: Statistically significant

These are the bare minimums. Most patterns you see exceed these thresholds.

High-Quality Patterns

|r| ≥ 0.6: Strong to very strong correlation
n ≥ 100: Large, robust sample
p < 0.01: Highly significant

When you see these numbers, pay attention. These are well-tested, reliable patterns.

Elite Patterns

|r| ≥ 0.7: Very strong correlation
n ≥ 150: Very large sample
p < 0.001: Extremely significant

Rare but powerful. These are the highest-conviction setups in the system.

| Quality Tier | |r| Threshold | Sample Size | p-value | Signal Interpretation | |--------------|---------------|-------------|---------|----------------------| | 🥉 Minimum | ≥ 0.3 | ≥ 30 | < 0.05 | Emerging pattern - use with caution and additional confirmation | | 🥈 High-Quality | ≥ 0.6 | ≥ 100 | < 0.01 | Reliable pattern - well-tested across many scenarios | | 🥇 Elite | ≥ 0.7 | ≥ 150 | < 0.001 | Highest conviction - rare but powerful setups |

When evaluating drivers, match them against these tiers to quickly assess quality. Elite patterns deserve the most attention in your analysis.

Common Mistakes Interpreting R-Values

Mistake 1: Assuming Causation

Wrong interpretation: "r=0.70 means insider buying causes the stock to rise"

Correct interpretation: "r=0.70 means historically, when insiders bought, stocks tended to rise. This could be because insiders have information, or because they buy when stocks are undervalued, or both."

Correlation ≠ causation. The pattern is real, but the mechanism might be complex.

Learn more: Correlation vs Causation

Mistake 2: Ignoring Sample Size

Wrong interpretation: "r=0.85 is amazing, I'm going all-in"

Missing context: What if n=12? That could be luck. Always check sample size.

Correct interpretation: "r=0.85 with n=200 is amazing. r=0.85 with n=12 needs more data."

Mistake 3: Comparing Apples to Oranges

Wrong comparison: "Pattern A has r=0.50 and Pattern B has r=0.65, so B is better"

Missing context:

What are the time horizons? (5-day vs 30-day returns)
What are the sample sizes?
What are the sectors?
What are the market conditions?

Correct approach: Compare patterns with similar characteristics (time horizon, sector, sample size) or evaluate each pattern independently against your strategy requirements.

Mistake 4: Treating R-Values as Probabilities

Wrong interpretation: "r=0.70 means 70% win rate"

Correct interpretation: "r=0.70 measures how consistently two variables move together, not win rate. Win rate requires different analysis (success rate in backtests)."

R-value and success rate are related but different metrics. Check both.

R-Values Across Different Time Horizons

Correlations change based on the time period measured:

Short-Term Patterns (5-10 days)

Typical r-values: 0.35 - 0.55
Why lower: More noise in short-term data
Best for: Day traders, swing traders
Require: Higher minimum r-value (≥0.4) for reliability

Medium-Term Patterns (10-30 days)

Typical r-values: 0.45 - 0.65
Why higher: Noise smooths out over time
Best for: Swing traders, position traders
Sweet spot: Most reliable timeframe for correlation analysis

Long-Term Patterns (30+ days)

Typical r-values: 0.40 - 0.70
Why variable: Many factors influence long-term returns
Best for: Position traders, investors
Require: Larger sample sizes to validate

Practical Application: Using R-Values

Step 1: Filter by Minimum Strength

Only consider drivers with |r| ≥ 0.5 for your primary signals. Use 0.3-0.5 range as confirmation only.

Step 2: Check Sample Size

Verify n ≥ 50 (preferably n ≥ 100) before trusting the correlation.

Step 3: Evaluate Direction

Positive r: Momentum or trend-following strategy
Negative r: Mean reversion or contrarian strategy

Match the correlation direction to your trading style.

Step 4: Combine Multiple Drivers

Look for 2-3 drivers with:

Similar r-values (all strong)
Same direction (all positive or all negative)
Different types (insider + technical + fundamental)

When multiple strong correlations align, confidence increases exponentially.

Real-World Example: Evaluating a Pattern

Pattern Details

    
    Driver: net_insider_buying_20d → forward_return_30d
r = 0.68
n = 176
p < 0.001
Contribution: 32%

Step-by-Step Evaluation

1. Check R-Value Strength

r = 0.68 falls in "strong positive" range (0.5-0.7)
✅ Above our 0.5 threshold for high-quality patterns

2. Verify Sample Size

n = 176 is well above our 100+ target for robust samples
✅ Highly confident this isn't luck

3. Confirm Significance

p < 0.001 means less than 0.1% chance this is random
✅ Statistically very significant

4. Assess Contribution

32% contribution means this driver heavily influences the score
✅ Primary driver, not minor contributor

Conclusion

This is an elite pattern. All factors align:

Strong correlation (0.68)
Large sample (176)
Highly significant (p < 0.001)
Major contributor (32%)

Trading implication: When this driver appears in an opportunity, give it significant weight in your analysis.

Common Questions

Q: What's the minimum r-value I should trade?

A: It depends on your strategy, but we recommend |r| ≥ 0.5 for primary signals and |r| ≥ 0.3 only as confirmation. Lower correlations are too noisy for reliable trading.

Q: Can r-values change over time?

A: Absolutely. Market regimes change, patterns decay, and correlations evolve. We continuously revalidate patterns and update r-values as new data arrives.

Q: Is r=0.9 always better than r=0.7?

A: Not necessarily. A pattern with r=0.9 and n=20 is less reliable than r=0.7 and n=200. Always consider sample size and statistical significance alongside r-value.

Q: Why do you show negative correlations?

A: Negative correlations are just as valuable as positive ones. They identify mean reversion opportunities, contrarian setups, and fade signals. Direction matters less than strength.

Q: How do I find the r-value for a specific opportunity?

A: Click into the opportunity's detail panel. The top drivers section shows the r-value, sample size, and contribution for each pattern.

Next Steps

Now that you understand correlation strength, learn why Sample Size Importance is equally critical for evaluating pattern quality.

Ready for deeper statistical concepts? Explore Statistical Significance to understand p-values and confidence intervals.

Understanding Correlation Strength (R-Values)

Documentation

Need help?