Prof. Melvin Hinich passed away seven months ago on Sept 7, 2010 (click here for my first post). I lost a great mentor whom I can always refer to whenever I have any questions on mathematics/statistics/econometrics. I have kept most of our email correspondences, all with his classic Hinich sign-off: “Be well, Mel”.
In memoriam of Melvin Hinich, Ordershook et al. (2011) and Munger et al. (2011) summarized Mel’s path-breaking contributions to academia over the past four decades (to download the papers, click here and here). Though Mel achieved international reputation in four academic disciplines (economics, engineering, political science and statistics), he will be most remembered for his seminal contributions to the use of spatial anaysis in analyzing politics and public policy. Because of that, as highlighted by the above two papers, spatial voting choice theory occupies a central role in political science just as how fundamental consumer choice is to economics.
It is indeed a great honour for me to work with such a reputable scholar, who has made a huge impact in academia. In my eight years of collaboration with Mel, I can bear testimony to the descriptions in Ordershook et al. (2011: 1) and Munger et al. (2011:160):
“… with an endless energy for work and a childlike curiosity about the science of almost everything. Hinich’s scholarship blended technical virtuosity, theoretical depth, interdisciplinary sweep, and a keen eye for the main chance in terms of substantive importance. But Mel was not simply a bright but easily distracted scholar with many interests. Rather, he was a scientist, a scholar who found most problems interesting, and he was capable of making connections across fields because so many problems share a deep logical and mathematical structure.”
In time series analysis, Mel advocated the importance of identifying nonlinear serial dependencies. According to Hinich and Patterson (1985), many early investigators implicitly assume the observed time series is generated from a Gaussian process and test for white noise using the correlation structure, hence ignoring possible nonlinear relationships between consecutive price changes. From a statistical perspective, the distinction between white noise and pure white noise is nontrivial when nonlinear dependence is present. With that conviction, he developed nonlinearity tests using the bispectrum (Hinich, 1982), and later extended to the next polyspectral measure, the trispectrum (Dalle Molle and Hinich, 1995). To complement the frequency-domain, Mel has contributed the time-domain counterparts in bicorrelation test (Hinich, 1996), cross-bicorrelation test (Brooks and Hinich, 1999), and tricorrelation test (Wild et al., 2010). With the development of powerful statistical tools, the literature witnessed a surge of empirical evidence supporting the presence of nonlinearity in stock market data, with Hinich and Patterson (1985) the first one (see the extensive literature survey by Lim and Brooks, 2011).
The existence of nonlinearity calls into question the adequacy of linear models, and hence invites the development of non-linear time series models which are expected to provide superior forecasts than their linear counterparts or the naïve random walk. However, the evidence to date on the out-of-sample forecasting performance of non-linear time series models is still unconvincing. The title of Ramsey (1996) rightly pointed out the implication: “If nonlinear models cannot forecast, what use are they?” As typical of Mel’s character, he did not see the development of statistical tests as an endpoint, but how useful they are in explaining real world phenomena. Mel argued that the inability of researchers to make meaningful point forecasts of stock returns despite strong evidence of nonlinearity is caused by the episodic transient nature of such dependencies. His conjecture received wide empirical support across different financial markets. In fact, there is now a growing literature on time-varying predictability, with its theoretical foundation in the Adaptive Markets Hypothesis (see again the extensive literature survey by Lim and Brooks, 2011).
At the age of 71 years old, Mel still had an endless energy for research. As documented by Munger et al. (2011: 164): “On the day before he died, Mel talked with Bob Molyneux about their next visit and called up Munger to talk about a new chapter for the revised edition of Analytical Politics”. In fact, a few weeks before he died, he worked with his collaborators in Italy, Australia and Chile. His passion for research will always be my learning model.