Research

The disposition effect requires a reference price to classify a position as a gain or a loss, and for multi-purchase positions the literature defaults to the volume-weighted average price (VWAP). Using population-level data on household investors, we show the relevant reference depends on how the position was built: investors who buy below their running cost anchor on the last purchase price, while investors who buy above anchor on the VWAP, consistent with Thaler's hedonic editing. The correction reverses two canonical findings: VWAP masks a 249-basis-point welfare cost, and experienced investors retain rather than shed their bias.

How does investor agreement at the 52-week high affect market quality? Using social media sentiment, intraday TAQ data, and investor-class order flows for all S&P 500 stocks, we document consensus: both retail and institutional investors predominantly sell at the 52-week high. The consensus tightens spreads by 15 basis points and increases sell-side liquidity by 16%, while informed trading falls by 30%. The high creates a liquidity barrier that reduces subsequent five-day returns by 30 basis points, challenging the conventional wisdom that high liquidity universally improves market quality.

Manufacturing firms face substantial losses from rare supplier quality failures, yet lack accessible tools to predict them. We develop a four-stage AI pipeline that integrates large language models into machine learning workflows, automating feature engineering from raw supplier data, unsupervised risk segmentation, ensemble model configuration for imbalanced prediction, and interpretability analysis with actionable recommendations. Validation on 622,924 aerospace transactions with class imbalance up to 139:1 shows significant improvements in F1, ROC-AUC, and PR-AUC over manual approaches, enabling organizations without data science expertise to deploy predictive supplier risk detection.

Cities differ in their openness to adopting new innovative products, and that variation shapes local firm value creation. We construct a proxy for city-level openness from the likelihood that new music is first played by local radio stations. Openness exhibits persistent cross-sectional variation across U.S. cities traceable more than a century, and during 2000 to 2019 it explains variation in the success of new ventures and new product introductions.