Bayesian Inference for Trading — Updating Beliefs with Smart Money Data

Bayesian reasoning is ideal for trading: you have a prior belief about price direction, then update it with each new piece of evidence (on-chain data, whale activity, funding rates). Smart Money API provides exactly this kind of evidence—update your beliefs and make probabilistic decisions.

Key concept: Bayes' Theorem: P(B|A) = P(A|B) × P(B) / P(A). In trading: P(Price↑ | Whales↑) = how likely is price up given that whales are accumulating?

Bayes' Theorem in Trading

The Formula

Prior: P(Direction) = your initial belief (60% chance BTC goes up)
Likelihood: P(Evidence | Direction) = probability you'd see this evidence if direction is true
Evidence: P(Evidence) = base rate of seeing this evidence
Posterior: P(Direction | Evidence) = updated belief after observing evidence

Example: Whale Accumulation as Evidence

Prior: 50% chance BTC goes up (baseline). Likelihood: If BTC goes up, there's a 72% chance whales accumulate (they front-run moves). If BTC stays flat/down, only 20% chance of accumulation. Evidence: You observe 250+ whale wallets accumulating.

Python — Bayesian update
def bayesian_update(prior_up, likelihood_evidence_given_up, likelihood_evidence_given_down):
# Prior
p_up = prior_up  # 0.50
p_down = 1 - prior_up  # 0.50
# Evidence likelihood
p_evidence_given_up = likelihood_evidence_given_up  # 0.72
p_evidence_given_down = likelihood_evidence_given_down  # 0.20
# Total probability of evidence (law of total probability)
p_evidence = (p_evidence_given_up * p_up) + (p_evidence_given_down * p_down)
# = (0.72 * 0.50) + (0.20 * 0.50) = 0.46
# Posterior: P(Up | Evidence) = Bayes
posterior_up = (p_evidence_given_up * p_up) / p_evidence
# = (0.72 * 0.50) / 0.46 = 0.783 (78.3% likely to go up)
return posterior_up

Result: After observing whale accumulation, probability of up moves from 50% to 78%. That's a strong signal to go long.

Multi-Signal Bayesian Framework

Combine multiple pieces of evidence (Smart Money API gives you three: derivatives, on-chain, whales):

Python — Multi-signal Bayesian
def multi_signal_bayesian(prior, deriv_signal, onchain_signal, whale_signal):
# Sequential updating: start with prior, update with each signal
posterior = prior
# Signal 1: Derivatives (funding rate, LSR)
posterior = bayesian_update(posterior, 0.68, 0.35)  # if up, 68% chance pos FR
# Signal 2: On-chain (MVRV, SOPR)
posterior = bayesian_update(posterior, 0.65, 0.40)  # if up, 65% chance MVRV favorable
# Signal 3: Whales (consensus, PnL)
posterior = bayesian_update(posterior, 0.72, 0.20)  # if up, 72% chance whale consensus long
return posterior

Likelihood Estimation from Historical Data

How do you know P(Evidence | Direction)? Calculate from backtests:

Python — Calibrate likelihoods
def estimate_likelihoods(historical_data):
# Of all times whales went long, what % saw price go up next 4h?
whale_long_ups = len(historical_data[(historical_data['whale_long'] == True) & (historical_data['next_4h_up'] == True)])
whale_long_total = len(historical_data[historical_data['whale_long'] == True])
p_up_given_whale_long = whale_long_ups / whale_long_total  # 0.62
# Of all times whales went short, what % saw price go down?
whale_short_downs = len(historical_data[(historical_data['whale_long'] == False) & (historical_data['next_4h_up'] == False)])
whale_short_total = len(historical_data[historical_data['whale_long'] == False])
p_down_given_whale_short = whale_short_downs / whale_short_total  # 0.58
return {'p_up_given_whale_long': p_up_given_whale_long, 'p_down_given_whale_short': p_down_given_whale_short}

Betting on Posterior Probabilities

Once you have a posterior probability, size your bet proportionally:

Python — Kelly Criterion from posterior
def kelly_bet_size(posterior_prob, odds=1.1):
# Kelly Criterion: f = (b*p - q) / b
# p = win probability, q = loss prob, b = odds ratio
p = posterior_prob
q = 1 - p
b = odds - 1  # if you win, you get 1.1x back (10% profit)
kelly_frac = (b * p - q) / b
# Use 25% of Kelly to be conservative
position_size = kelly_frac * 0.25
return max(0, position_size)  # never negative
# Example: posterior = 78%, Kelly = 6.5%, use 1.6% of account

Conjugate Priors for Efficiency

For continuous estimates (e.g., "what's the true win rate of whales?"), use Beta priors:

Beta(α=1, β=1) = uniform (no prior knowledge)
Beta(α=10, β=5) = believes whales win ~67% but uncertain
Beta(α=100, β=50) = strong belief in 67% win rate

After observing N wins and M losses, posterior is Beta(α + N, β + M). This conjugate structure lets you update instantly without sampling.

Make probabilistic decisions with Smart Money

Our API returns composite scores and confidence levels—the exact inputs for Bayesian frameworks. Build a trading system that updates beliefs with whale activity, on-chain metrics, and derivatives signals.

Learn Bayesian Trading →