Research
Research
Hong Kong Institute for Monetary and Financial Research (HKIMR) Research Paper No. 12/2026
Selected Presentations: AFA Poster (scheduled), SFS Cavalcade Asia-Pacific (scheduled), Global AI Finance Research Conference (scheduled), UC Berkeley, AsianFA Doctoral Consortium, Durham Conference for Finance Job Market Paper, FMA Annual Meeting, Hong Kong Monetary Authority, Edinburgh Financial Technology Conference, FMARC Doctoral Consortium, CES China Annual Conference, Gillmore Center Fintech PhD Workshop
Abstract
Large language models can improve investors' understanding of disclosures while making their errors more correlated. I study this trade-off around ChatGPT's public launch using pre-launch disclosure readability and uncertainty. Within stocks, trading becomes more synchronized: order flow turns more one-sided in the hardest-to-read decile than in the easiest, by 6.4 percent of the sample mean. Across stocks, the order flow of hard-to-read firms moves more with that of other stocks, by 13 percent of the pre-launch mean per standard deviation of readability difficulty. Price discovery also improves: relative to comparable firms, a larger share of the price impact of order flow becomes permanent, and earnings news enters prices faster. Greater disclosure uncertainty predicts less one-sided trading and no improvement in price discovery. A Kyle-style model accounts for these patterns in a single framework. Investors can delegate reading to a common language model that reduces decoding error, leaves uncertainty about outcomes unresolved, and introduces shared error. If every investor had to delegate, prices could become less informative even as each reads more precisely. When adoption is a choice, crowding limits it, and greater synchronization coexists with better price discovery.
Abstract
This paper develops a game-theoretic model of endogenous liquidity fragmentation in decentralized exchanges with automated market makers. Liquidity providers strategically create competing pools, while traders optimally split orders across pools by balancing price impact and gas fees. When gas fees are sufficiently small, no pure-strategy equilibrium exists because entry by additional pools is always profitable. For intermediate gas fees, multiple fragmented equilibria may coexist. For sufficiently large gas fees, fragmentation collapses and the market converges to a unique one-pool equilibrium. We then show that the presence of front-running transactions on decentralized exchanges creates additional motives for fragmentation: by splitting orders across pools, traders reduce rents extractable through front-running transactions. The equilibrium degree of fragmentation is therefore endogenously determined by the interaction among gas fees, entry incentives, and front-running risk.
Minor revision, Journal of Corporate Finance
(Draft available upon request)
Abstract
In this paper, we propose a unified model framework to directly derive a firm-specific measure of the expected weighted average cost of capital. Our results provide evidence of a U-shaped (inverted U-shaped) relationship between the cost of capital (the market value of the firm) and debt usage in unregulated industries, supporting the trade-off theory at the industry level. The CFOs’ preferred debt usage metric, Debt/EBITDA, performs as well as the market value debt-to-asset ratio and better than the book value ratio. We also find that firms with a low cost of capital invest more, and that firms with high unlevered beta have a high cost of capital.
(Draft available upon request)
Abstract
Practically used price multiples are largely disconnected from theoretically sound discounted cash flow valuation models. In this paper, we introduce a ‘synthetic price-to-forward earnings multiple’ that utilizes a firm’s economic and accounting fundamentals, as well as key drivers of value in construction. We propose an approach to estimate the relevant parameters simultaneously. The synthetic price multiple provides an additional tool for valuing stocks and predicting stock returns. We find that the value estimates from our synthetic price-to-forward earnings multiples are less biased than the often-used valuation models and price multiples. The difference between synthetic and actual price-to-forward earnings multiples has important implications for equity investments. Longing firms with the lowest actual-to-synthetic price multiples and shorting those with the highest gaps can generate statistically and economically significant hedge returns.
(Draft available upon request)
Abstract
We model investors’ decisions to use coarse thinking to process information and the effects of this on asset markets. Agents with weak priors may choose to pool together their signals across assets to mitigate the problem of overfitting. This model of coarse information processing helps explain features of return comovement in markets that are hard to explain with existing models based on rational expectations or limited attention. Theoretically and empirically, we demonstrate a link between return comovement and return predictability.