Answers are grounded in this paper and cite their sources.
Let M = \| mᵢⱼ \| be a 4 × 4 irreducible aperiodic Markov matrix such that h₁ ≠ h₂, h₃ ≠ h₄, where hᵢ = mᵢ₁ + mᵢ₂. Let x₁, x₂, ⋯ be a stationary Markov process with transition matrix M, and let yₙ = 0 when xₙ = 1 or 2, yₙ = 1 when xₙ = 3 or 4. For any finite sequence s = (ε₁, ε₂, ⋯, εₙ) of 0's and 1's, let p(s) = Pr₁ = ε₁, ⋯, yₙ = εₙ\. If {equation*}{1}p^2(00) ≠ p(0)p(000) and p^2(01) ≠ p(1)p(010),{equation*} the joint distribution of y₁, y₂, ⋯ is uniquely determined by the eight probabilities $p(0), p(00), p(000), p(010), p(0000), p(0010), p(0100), p(0110)$, so that two matrices M determine the same joint distribution of y₁, y₂, ⋯ whenever the eight probabilities listed agree, provided (1) is satisfied. The method consists in showing that the function p satisfies the recurrence relation {equation*}{2}p(s, ε, δ, 0) = p(s, ε, 0)a(ε, δ) + p(s, ε)b(ε, δ){equation*} for all s and ε = 0 or 1, δ = 0 or 1, where a(ε, δ), b(ε, δ) are (easily computed) functions of M, and noting that, if (1) is satisfied, a(ε, δ) and b(ε, δ) are determined by the eight probabilities listed. The class of doubly stochastic matrices yielding the same joint distribution for y₁, y₂, ⋯ is described somewhat more explicitly, and the case of a larger number of states is considered briefly.
No takes yet. Share an insight, caveat, or question.
Blackwell et al. (1957) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: