arXiv subjects

Tomas Havranek

Publications and source records attributed to Tomas Havranek.

5 recordsLinked to original sources

arXiv2026

Optimal Inflation Rate: A Meta-Analysis

We revisit the optimal long-run inflation rate using 777 estimates from 116 primary studies published between 1989 and 2026, the largest sample on the topic to date. To our knowledge, this is among the first economics meta-analyses in which primary-data extraction is done from start to finish through a documented and auditable large-language-model pipeline, calibrated against a hand-coded training set and released for replication. The literature points to an optimum of about 0.6 percentage points per year, well below the two-percent targets used by most advanced-economy central banks. The gap should not be automatically read as a verdict against the two-percent norm. Measurement error in published price indices could close, widen, or even reverse the gap, and the structural literature itself cannot pin down the sign of the required correction. Bayesian model averaging over the full set of structural moderators shows that cross-study variation is driven by real modelling choices rather than by selective reporting. The main drivers are the choice of monetary benchmark (Friedman rule vs. laissez-faire), the transactions-frictions technology, the assumed shock structure, and the class of nominal-rigidity contract. The non-parametric caliper test finds no upward bunching at the two-percent target. The paper contributes a reproducible LLM-assisted extraction pipeline for structurally calibrated literature and a quantitative decomposition of where the optimal-inflation literature disagrees.

econ.GN↗

arXiv2026

Publication bias and p-hacking in the effect of COVID-19 on learning

We revisit a central estimate in the economics of education: the human-capital loss associated with COVID-19 school closures. Estimates of pandemic learning loss may be affected by publication bias, p-hacking, and the mechanical correlation between standardized effect sizes and their standard errors. We conduct a comprehensive multi-method assessment of bias by applying a wide range of correction techniques - including PET-PEESE, three-parameter selection models (3PSM), Robust Bayesian Meta-Analysis (RoBMA), Meta-Analysis Instrumental Variable Estimation (MAIVE), Right-Truncated Meta-Analysis (RTMA), and multi-bias sensitivity analysis. Our preferred specifications, RoBMA and MAIVE, rely on different assumptions yet converge on an effect size of approximately -0.12 SD, equivalent to a learning loss of about 30% of a school year. Although some methods reveal signs of publication bias and selective reporting, these findings do not explain away the central finding: the COVID-19 learning deficit is economically meaningful and statistically robust.

econ.GN↗

arXiv2026

Do decisions about outliers and influential effects matter? Evidence from 358 behavioral science meta-analyses

Meta-analysts routinely face estimates that look too large or extreme. Yet, how to handle them is left to the reviewer's judgment. The methods for detecting such estimates are well known. What is missing is an informed assessment of how much alternative handling choices might change a meta-analysis' conclusions. We fill this gap by analyzing the effects of four pre-registered handling treatments across 358 behavioral science meta-analyses with at least ten estimates. Each outlier handling treatment is estimated by two estimators (random effects and unrestricted weighted least squares), and compared to the 'do-nothing' baseline on three outcomes: the pooled effect, statistical significance, and whether the effect reaches the smallest effect size of interest (|d| >= 0.20). Our entire analysis and comparison pipelines were pre-registered. Alternative outlier handling treatments have little effect on the meta-analysis mean as the median absolute change in Cohen's d is at most 0.047 and often much less. Yet, at least one of these four treatments in combination with one of these estimators reverses the statistical significance of 11.5% of meta-analyses and the smallest-effect-of-interest assessment in 15.9%. Winsorizing has the least effect and DFBETAS the most. Categorical changes are found almost entirely among results already close to the decision boundary; strongly significant results essentially never change. These findings give applied meta-analysts, methods specialists, and reviewers a reference point for how much this under-reported choice matters and provide yet another reason for meta-analysts to publicly pre-specify their methods and handling treatments.

econ.EM↗

arXiv2026

Electricity demand has not become more price-responsive despite ninety years of technological change

Energy planners have long assumed that electricity demand will grow more price-responsive as metering, automation, and storage spread, an assumption now embedded in decarbonization plans. We test it against the empirical record: 4,720 own-price elasticity estimates from 462 studies, with data spanning 1934-2024, ranked on a single ladder of identification quality from naive regressions to randomized experiments. Three findings emerge. First, the best-identified studies find smaller responses than naive ones: the publication-bias-corrected short-run elasticity is about -0.16 (a 10% rise in the electricity price cuts consumption by under 2%), and only -0.09 among the best-identified studies, whose adjusted value is statistically indistinguishable from zero. Second, responsiveness grows with time to adjust, roughly doubling from -0.16 in the short run to -0.38 in the long run as the capital stock turns over, but this pattern has itself been stable for decades. Third, and most important, responsiveness shows no upward trend across nine decades of data; if anything, the most technology-rich settings, including time-of-use pricing, are the least price-responsive in total consumption. Prices alone have not made total electricity consumption more responsive; broader demand flexibility will have to be engineered and paid for, through enabling technology, contracts, and program design.

econ.GN↗

arXiv2026

Does Multi-Agent Debate Improve AI Feedback on Research Papers?

Probably not, at least for meta-analyses in economics. In a pre-registered, identity-masked, within-paper experiment, the authors of 44 meta-analyses ranked three AI reports on their own paper by usefulness for improving it: a single pass by a frontier model against two multi-agent debate tools we built and expected to win. All reports were held to a common length and template. The authors preferred the single pass, by 0.66 rank points over mad-research (95% CI 0.32 to 1.00) and 0.57 over paper-workshop (0.16 to 0.95), though paper-workshop spent roughly thirty times the tokens. Authors who recalled their journal referee report usually placed it first and never last; in a separate exercise, three AI judges almost always placed the real journal referee report last. Among the three AI reports, Gemini (the judge whose model family wrote none of the reports) would have ranked paper-workshop first in the authors' place, reversing the single-pass preference. The reversal warns against substituting an AI judge for the author. We measure perceived usefulness for finished papers; whether AI should referee papers is a separate question.

econ.GN↗