Ingvar Strid

Overview

Ingvar Strid is a Swedish economist and computational statistician associated with the Sveriges Riksbank. He developed the prefetching approach to parallelising Metropolis-Hastings algorithms, which speculatively evaluates the posterior at multiple future candidate states in parallel via a binary Metropolis tree, eliminating the serial bottleneck in RWMH. His key theoretical result is that the classical 0.234 optimal acceptance rate no longer holds in parallel settings: the true optimum decreases monotonically with the number of processors, reaching 0.120 for 7 processors and approaching zero as $P \to \infty$.

Key Contributions

• Prefetching for MH parallelisation (Strid 2010, Computational Statistics & Data Analysis 54: 2814–2835): introduced the Metropolis tree framework; derived $D(P)$ (expected draws per tour); specified five algorithm variants from basic (Brockwell 2006) to "most likely path" using a Laplace surrogate $p^* = \mathcal{N}(\theta_m, H_m^{-1})$; proved the downward shift in $\alpha_\text{opt,P}$; benchmarked on Smets-Wouters (2003) DSGE model achieving theoretical speedup 4.99× and actual cluster speedup 5.15× with 8 processors.
• DSGE state-space estimation (Strid-Walentin 2008, Computational Economics 33: 277–304): efficient state-space methods for DSGE likelihood evaluation (co-authored with Karl Walentin).

Related

• Metropolis-Hastings Algorithm
• Strid (2010)