U.S. Politics from a multifractal perspective

This paper studies the long-range dependence and multifractal content of U.S. political time-series to gather a deeper understanding of sociophysic phenomena. Specifically, multifractal detrended fluctuation analysis (MF-DFA) is applied upon data in the context of (i) president approval (polls), (ii) president online attention (Google Trends) and (iii) election-win probabilities (prediction markets). All analyzed series are characterized by anti-persistence, which may be interpreted as a nervous and overreacting behavior. We further detect significant multifractality with true non-linear correlation remaining after correcting for spurious sources. Importance from understanding the multifractal behavior arises from the fact that all three data types are used in practice for the prediction of election outcomes. We further argue that variation in local persistence (as implied by multifractality) can be both beneficial and destructive in different real-world scenarios. We draw parallels to simple examples like the timing of political campaigns or trading on prediction markets. On the methodological side, the article implements recent improvements of MF-DFA such as focus-based regression and overlapping segments.