Ergodic Probability

The ergodic probability vector π satisfies the balance equation π = πP, where P is the transition probability matrix. It represents the fixed point of the system's dynamics.

• Computed by solving the eigenvector equation π(I - P) = 0 with the normalization constraint Σπᵢ = 1
• For a two-regime model, the closed-form solution is π₁ = p₂₁/(p₁₂ + p₂₁) where pᵢⱼ is the probability of transitioning from regime i to j
• Represents the unconditional probability of being in a given state at any randomly chosen time in the distant future

A defining property of ergodic Markov chains is that the influence of the starting regime decays exponentially over time. After a sufficient mixing period, the probability of being in any state converges to the ergodic distribution.

• The mixing time quantifies how many steps are required for the state distribution to be within ε of the stationary distribution
• Mathematically: limₙ→∞ Pⁿ = 1π where each row of the limiting matrix equals the ergodic vector
• This property ensures that long-horizon regime forecasts are independent of current market conditions

The ergodic probability is directly linked to the expected sojourn time in each regime. The expected duration of regime i is E[Dᵢ] = 1/(1 - pᵢᵢ), where pᵢᵢ is the self-transition probability.

• The ergodic probability can be expressed as πᵢ = E[Dᵢ] / Σⱼ E[Dⱼ], meaning it equals the expected duration in state i divided by the total expected cycle length
• A regime with high persistence (pᵢᵢ close to 1) will dominate the ergodic distribution
• In financial applications, bull markets typically exhibit longer expected durations than bear markets, leading to asymmetric ergodic weights

For ergodic probabilities to exist and be unique, the Markov chain must satisfy two conditions: irreducibility and aperiodicity.

• Irreducibility: Every state must be reachable from every other state with positive probability in some number of steps. No absorbing or transient states are permitted
• Aperiodicity: The return times to any state must not be restricted to multiples of some integer greater than 1, preventing deterministic cycling
• In financial regime models, these conditions are typically satisfied because markets can transition from any regime to any other, and transitions occur at irregular intervals

The ergodic probability provides the strategic allocation benchmark for investors with infinite horizons. It answers the question: 'What fraction of time should I expect to spend in each market regime?'

• A regime-switching model with ergodic probabilities π_bull = 0.72 and π_bear = 0.28 implies the market spends approximately 72% of time in bull conditions
• Portfolio weights can be computed as the ergodic-weighted average of regime-conditional optimal allocations: w = Σᵢ πᵢ wᵢ**
• This contrasts with myopic, single-period optimization that ignores regime dynamics entirely
• The ergodic Sharpe ratio provides a more honest assessment of strategy performance than conditional metrics computed only in favorable regimes

In practice, verifying convergence to the ergodic distribution requires monitoring the total variation distance between successive iterates of the state probability vector.

• The second-largest eigenvalue (λ₂) of the transition matrix governs the rate of convergence; smaller magnitudes imply faster mixing
• The spectral gap defined as 1 - |λ₂| quantifies how quickly the chain forgets its initial conditions
• For high-frequency trading applications, if the mixing time exceeds the strategy horizon, the ergodic assumption is invalid and conditional state probabilities must be used instead
• Gelman-Rubin diagnostics from Bayesian inference can be adapted to assess whether multiple simulated chains have converged to the same stationary distribution