P-Value
Published Last updated
Key Takeaway
Probability statistic indicating likelihood that observed cryptocurrency price relationships occurred by random chance alone, guiding decisions to accept or reject trading strategy hypotheses.
What Is P-Value?
Probability statistic indicating likelihood that observed cryptocurrency price relationships occurred by random chance alone, guiding decisions to accept or reject trading strategy hypotheses.
How P-Value Works
Frequently Asked Questions
What does a p-value tell me about my cryptocurrency trading relationship?
A p-value quantifies probability that observed patterns occurred randomly without genuine relationships. If your Augmented Dickey-Fuller test produces p-value of 0.03, you can conclude with 97 percent confidence the relationship shows mean-reverting stationarity (only 3 percent probability of random occurrence). P-value of 0.45 suggests 45 percent random probability—insufficient evidence to conclude stationarity; treat the pair as non-stationary. Standard trading threshold is p < 0.05: below this threshold, statistical evidence supports the relationship; above it, the relationship appears random. Remember: low p-values suggest genuine patterns; high p-values suggest coincidences.
Why doesn't a low p-value guarantee my cryptocurrency trading strategy will profit?
Low p-values confirm statistical significance—relationships are probably real, not random. However, statistically significant relationships may be too weak to exceed trading costs. Two assets might have p-value 0.01 (highly significant stationarity) yet their spread reverts so slowly (extended half-life) that profits fail to cover trading fees and slippage. Additionally, in-sample p-values suffer from overfitting: you can find statistically significant relationships by testing enough candidate pairs; most are coincidences. Professional validation requires: (1) low p-values confirming statistical significance, (2) reasonable effect size (profitable spread movements), (3) out-of-sample validation confirming in-sample p-values predict future performance. All three conditions must hold.
How do I avoid false-positive p-values when screening many cryptocurrency trading pairs?
When testing hundreds of candidate pairs for stationarity, random probability guarantees that approximately five percent will show p-values below 0.05 purely by chance, even without genuine relationships. This multiple-testing problem inflates false positives significantly. Solutions: (1) Tighten p-value thresholds (require p < 0.01 instead of p < 0.05) reducing false positives, (2) Use Bonferroni correction adjusting thresholds based on number of tests performed, (3) Validate candidate pairs through out-of-sample testing before deployment, (4) Cross-validate findings on independent data. Professional quant traders employ conservative p-value standards (p < 0.01) and mandatory out-of-sample validation when screening many candidates.
Common Misconceptions About P-Value
A p-value of 0.001 means my cryptocurrency trading relationship is one thousand times more significant than a p-value of 0.001 × 1000 = 1.0.
This mathematical interpretation is incorrect and dangerous. P-values aren't directly comparable as importance measures. A p-value of 0.001 versus 0.05 indicates stronger evidence against randomness but doesn't quantify relationship strength, profitability, or trading viability. Two pairs might have identical p-values yet very different practical significance: one produces consistent profits, another barely covers costs. Effect size (spread reversion speed, volatility) determines trading profitability, not p-values. Use p-values as initial statistical filters, then assess effect size and profitability separately.
If my trading hypothesis shows p-value 0.04, I can deploy my strategy confidently knowing it has 96 percent success probability.
This misinterprets p-value fundamentally. P-value 0.04 means 4 percent probability the observed results occurred randomly IF the null hypothesis is true (no real relationship). It does NOT mean 96 percent probability of future trading success. The hypothesis might be real (4 percent random probability) yet still unprofitable after trading costs, or dependent on market regime changes. Success probability requires out-of-sample validation, not p-value interpretation. Traders confusing these concepts deploy strategies with low p-values that fail in live trading, suffering capital losses from misunderstood statistics.
P-values measure how strong my cryptocurrency relationship is or how much profit I'll make.
P-values measure random-chance probability alone; they measure neither relationship strength nor profitability. A mean-reversion spread might have p-value 0.02 (statistically significant) but revert over ninety days (too slow for profitable trading); another pair has p-value 0.03 with five-day reversion (highly tradeable). Effect size, trading costs, and execution time determine profitability; p-values don't address these factors. Use p-values as yes/no statistical filters (real versus random), then separately evaluate effect size and trading economics. Confusing p-value significance with practical significance is a primary source of trader losses.