Quandt 1958 1972

Richard E. Quandt’s two switching-regression papers are the historical headwaters of the regime-switching family. “The Estimation of the Parameters of a Linear Regression System Obeying Two Separate Regimes” (Journal of the American Statistical Association 53(284), 1958, 873-880) posed the original problem: when a process obeys one linear regression up to some unknown date and a different one afterwards, “it is necessary first to estimate the position of the point in time at which the switch from one regime to the other occurred”, and Quandt proposed a maximum-likelihood procedure “based upon a direct examination of the likelihood function”. “A New Approach to Estimating Switching Regressions” (JASA 67(338), 1972, 306-310) generalised this, and the switching-regression apparatus was subsequently applied to market-disequilibrium problems where supply and demand schedules cannot both be observed.

These papers introduced the time-varying-parameter idea — a regression whose coefficients differ across regimes — but the regime mechanism was a deterministic or independently drawn switch, not yet a stochastic process with memory. Kim, Piger and Startz (2008) place Quandt at the start of the lineage: “regime-switching regressions … date to at least Quandt (1958)“. The decisive next step was Goldfeld and Quandt 1973, which replaced the independent switch with a latent first-order Markov chain (giving the state serial persistence), after which Hamilton 1989 extended the Markov-switching model to autoregressive, serially dependent data. Hamilton’s Palgrave survey notes that Goldfeld & Quandt’s independent switching model (1972) and fixed-transition-probability Markov model (1973) are distinct points on this path, with Quandt (1972) corresponding to the time-varying-transition independent-switching variant.

For this vault these are pure method papers. They are descriptive/estimation statistics: they define an estimation problem and a maximum-likelihood solution, test no trading strategy, use no out-of-sample period, and report no return, Sharpe ratio, drawdown, or hit rate. They appear here only as the documented antecedents of the Markov Regime-Switching Model — the “ancestor” notes a reader should know exist when assessing where the model class came from. They carry no profitability evidence, hence profitability_evidence_grade: inconclusive.

Quandt 1958 1972 [precedes] Goldfeld and Quandt 1973 Quandt 1958 1972 [precedes] Markov Regime-Switching Model Goldfeld and Quandt 1973 [precedes] Hamilton 1989

Connections

Sources