第八届全球量化和宏观投资会议于2024年10月23日至24日在纽约野村证券办公室举行。此次会议包含了一系列由顶尖学者和专业人士带来的报告，涵盖了量化金融、宏观经济分析、人工智能和机器学习等多方面的主题。
本次会议首次设置了主题演讲，康奈尔大学的丛林教授（Lin William Cong）在会上介绍了AlphaManager框架。AlphaManager通过数据驱动的稳健控制（Data-Driven-Robust-Control, DDRC）方法，将人工智能整合到公司财务决策中。该系统由两个主要组成部分构成：环境预测模块（Predictive Environment Module, PEM）利用深度学习捕捉企业环境的动态；政策模块（Policy Module）则借助强化学习确定最优管理策略。AlphaManager方法能够有效地模拟复杂的经济系统，并优化管理决策，避免了高成本的实验。通过结合数据驱动的实证方法与人工智能优化技术，AlphaManager填补了公司财务研究中的关键空白。它的目标是提供更加全面且实用的经济环境描述，同时具备实时适应决策的能力。
以下是会议上展示的部分论文及其摘要：
We develop a unique dataset, the first-ever of its kind, by leveraging the US Census Bureau's LEHD program and various big textual data sources, to examine the factors influencing the compensation and career trajectories of US active equity mutual fund managers. We find that managers' compensation is primarily determined by assets under management (AUM), with return performance directly influencing bonuses beyond its impact on AUM. Despite not aligning with client interests, fund flows significantly affect manager compensation and career outcomes. Large fund outflows increase a manager's likelihood of job turnover (with a substantial decline in compensation) by 4 percentage points.
Companies’ strategic alternatives announcements lead to negative future stock returns. We first investigate whether an anomaly exists and demonstrate that the anomaly is significant, pervasive across years, industries, firm size, and information environments, and not driven by confounding variables nor risk. We then investigate why investors misprice the announcements and find that investors appear overly optimistic about a potential merger or acquisition and do not fully incorporate the negative fundamental news conveyed by the announcement. Meanwhile, short sellers exploit the mispricing. We also evaluate market frictions as limits to arbitrage. This study’s contributions are: (i) evaluating mispricing and risk explanations for an event that causes extreme stock returns, (ii) challenging investors’ widely held belief that such announcements reflect good news, and (iii) warning investors and analysts about a behavioural bias they might unknowingly adopt.
Using hand-collected data, we provide evidence on the information and monitoring roles of sell-side equity analysts who previously studied or worked in accounting. Relative to the average analyst, only former auditors issue more accurate EPS forecasts and more profitable sell recommendations. Firms covered by former auditors are less likely to report material misstatements, suggesting that they play a monitoring role. Analysts with greater prior auditing experience drive these results. In contrast, analysts with a university degree in accounting, a CPA certification, or prior corporate experience as an accountant do not outperform. Therefore, we conclude that the combined accounting and industry knowledge acquired during several years of work in public accounting can give former auditors a competitive advantage. Consistently, former auditors ask more accounting-related questions with a less positive tone than other analysts during earnings conference calls. Overall, our results highlight the extent and limits of accounting expertise in sell-side research.
Corporate decision-making involves high-dimensional, non-linear stochastic control under managerial learning and dynamic interactions with the economic environment. We introduce an AI-assisted, data-driven-robust-control (DDRC) framework to complement theory, reduced-form models, and structural estimations in corporate finance research. We do so with an emphasis on explaining and predicting firm outcomes empirically, and offering policy recommendations for any business objectives. Specifically, we build a predictive environment module through supervised neural networks and add a policy module through deep reinforcement learning that goes beyond hypothesis testing on historical data or simulations. By incorporating model ambiguity and robust control techniques, our framework not only better explains and predicts corporate outcomes in- and out-of-sample, but also identifies important managerial decisions while offering effective policy recommendations adaptive to market evolution and feedback. We also document the rich heterogeneity in model prediction performance, ambiguity, and policy efficacy in the cross section of U.S. public firms and across time regimes. Critically, our DDRC approach distinguishes between scenarios where theory and causal identification are important from situations where predictive models trained on historical observations suffice. It informs where corporate finance research should focus, allowing for the incorporation of fragmented knowledge through ambiguity-guided transfer learning.
Timely monetary policy decision-making requires timely core inflation measures. We create a new core inflation series that is explicitly designed to succeed at that goal. Precisely, we introduce the Assemblage Regression, a generalized nonnegative ridge regression problem that optimizes the price index's subcomponent weights such that the aggregate is maximally predictive of future headline inflation. Ordering subcomponents according to their rank in each period switches the algorithm to be learning supervised trimmed inflation — or, put differently, the maximally forward-looking summary statistic of the realized price changes distribution. In an extensive out-of-sample forecasting experiment for the US and the euro area, we find substantial improvements for signaling medium-term inflation developments in both the pre- and post-Covid years. Those coming from the supervised trimmed version are particularly striking, and are attributable to a highly asymmetric trimming which contrasts with conventional indicators. We also find that this metric was indicating first upward pressures on inflation as early as mid-2020 and quickly captured the turning point in 2022. We also consider extensions, like assembling inflation from geographical regions, trimmed temporal aggregation, and building core measures specialized for either upside or downside inflation risks.
This article introduces a novel hybrid regime identification-forecasting framework designed to enhance multi-asset portfolio construction by integrating asset-specific regime forecasts. Unlike traditional approaches that focus on broad economic regimes affecting the entire asset universe, our framework leverages both unsupervised and supervised learning to generate tailored regime forecasts for individual assets. Initially, we use the statistical jump model, a robust unsupervised regime identification model, to derive regime labels for historical periods, classifying them into bullish or bearish states based on features extracted from an asset return series. Following this, a supervised gradient-boosted decision tree classifier is trained to predict these regimes using a combination of asset-specific return features and cross-asset macro-features. We apply this framework individually to each asset in our universe. Subsequently, return and risk forecasts which incorporate these regime predictions are input into Markowitz mean-variance optimization to determine optimal asset allocation weights. We demonstrate the efficacy of our approach through an empirical study on a multi-asset portfolio comprising twelve risky assets, including global equity, bond, real estate, and commodity indexes spanning from 1991 to 2023. The results consistently show outperformance across various portfolio models, including minimum-variance, mean-variance, and naive-diversified portfolios, highlighting the advantages of integrating asset-specific regime forecasts into dynamic asset allocation.
We study probability forecasts in the context of cross-sectional asset pricing with a large number of firm characteristics. Empirically, we find that a simple probability forecast model can surprisingly perform as well as a sophisticated probability forecast model, and all of which deliver long-short portfolios whose Sharpe ratios are comparable to those of the widely used return forecasts. Moreover, we show that combining probability forecasts with return forecasts yields superior portfolio performance versus using each type of forecast individually, suggesting that probability forecasts provide valuable information beyond return forecasts for our understanding of the cross-section of stock returns.
We study the impact of foreign information on asset prices by applying machine learning to detect foreign signals that predict daily U.S. stock returns. Candidate foreign signals include lagged returns of stock markets and individual stocks in 47 foreign markets. Over 100,000 models are trained to capture stock-specific time-varying relationships between foreign signals and returns. Foreign signals exhibit out-of-sample return predictability for domestic and multinational companies and across industries. A portfolio formed on return forecasts by foreign signals generates abnormal returns of 5.77 basis points per day. Signal importance analysis reveals the price discovery of foreign information takes multiple weeks and is much slower for information from emerging and low-media-coverage markets and among stocks with lower foreign institutional ownership. Valuable foreign signals are not concentrated in those largest countries nor foreign firms in the same industry as U.S. firms. Our study suggests that machine learning-based investment strategies exploiting foreign signals can emerge as important mechanisms to facilitate foreign information dissemination into asset prices.
This paper presents a comparative analysis evaluating the accuracy of Large Language Models (LLMs) against traditional macro time series forecasting approaches. In recent times, LLMs have surged in popularity for forecasting due to their ability to capture intricate patterns in data and quickly adapt across very different domains. However, their effectiveness in forecasting macroeconomic time series data compared to conventional methods remains an area of interest. To address this, we conduct a rigorous evaluation of LLMs against traditional macro forecasting methods, using as common ground the FRED-MD database. Our findings provide valuable insights into the strengths and limitations of LLMs in forecasting macroeconomic time series, shedding light on their applicability in real-world scenarios.
Although researchers often view earnings management as being widespread, measuring the cost and level of earnings management is a non-trivial task. We derive a measure of earnings management cost and the associated equilibrium level of earnings management from the cross-sectional properties of earnings and prices. This approach enables us to separate economic shocks from reporting discretion by modeling the economic trade-off faced by management. The trade-off can be easily estimated from a closed-form likelihood function. Consistent with prior studies, the measure suggests more earnings management during seasoned equity offerings, for smaller and growing firms, as well as in industries with more irregularities.
Individual stocks with daily closing price just above round numbers (e.g., $6.1) outperform stocks just below (e.g, $5.9) by 24.6 bps over next day, which translates to an annualized return of 61%! This return pattern is extremely robust to various stock characteristics and international samples. We provide evidence to link these predictable price movements to limit order clustering: prices just above (below) round numbers are supported (resisted) by a clustering of limit orders at these psychologically significant levels. This finding reveals a profound impact of order clustering (mostly by retail investors) on random price movement and market efficiency.
We introduce a conditional machine learning (CML) approach to forecasting stock index returns under the assumption that the cross-section of expected stock returns has a latent factor structure. We estimate the factors from the cross section deep neural network-based forecasts of stocks returns, which we use to construct forecasts of stock index returns. A key contribution of our work is the derivation of closed-form expressions for the standard error/uncertainty of the CML forecast, which has significant economic implications. First, the derived forecast uncertainty enables the ensembling of CML with forecasts from standard machine learning models, resulting in a more stable ensemble forecast with fewer breaks in predictability. Second, the forecast uncertainty can be economically explained by the `CDI" index, defined as the correlation between firms' ranks based on sales revenue and their ranks based on market value of shares. Our analysis shows that periods of poor CML performance correspond to times when the CDI is relatively low, indicating stronger creative destruction forces and greater instability in the factor structure of expected returns, i.e., when younger firms are more of a threat to older firms. Additionally, during these times, the volatility of our conditional model's prediction errors is also more volatile.
We investigate the role of market microstructure metrics in explaining and predicting price dynamics for 5 major cryptocurrencies. Using machine learning, we show how microstructure measures of liquidity and price discovery have predictive power for price dynamics of interest for electronic market making, dynamic hedging strategies and volatility estimation. We identify important own market and cross-market effects for BTC and ETH Roll measures and VPINs. Our results are little changed during crypto winter, demonstrating a stability to these effects. Our findings suggest that market dynamics of cryptocurrencies can be viewed as similar to those of other investible asset classes.
This paper examines the role of peer firm disclosures in shaping corporate R&D investments. Drawing on models of two-stage R&D races, I hypothesize that a firm could be either deterred or encouraged by peer disclosure of interim R&D success, depending on peer firms’ R&D strength in the race. Using granular, project-level data on clinical trials in the drug development process, I find that a firm’s R&D investments in a specific therapeutic area are deterred by disclosures of early-phase trial initiation from strong rivals in the same area but encouraged by disclosures from weak rivals. Cross-sectional analyses show that focal firm strength and disclosure relevance moderate the effects of peer firm disclosure. Overall, my evidence suggests that peer firms’ R&D disclosures can have both proprietary costs and deterrence benefits.
Machine learning and AI technologies can identify data patterns related to financial misreporting that traditional methods overlook. This study investigates whether increasing machine readership of corporate financial statements influences managers' financial reporting decisions. Our findings indicate a reduction in financial misreporting when machine readership is higher, and these results are consistent after addressing potential identification issues. Notably, for misreporting patterns detectable by traditional linear models, we do not observe any incremental disciplining effect from machine readership, indicating that the strength of machines lies in recognizing non-linear and high-dimensional patterns. The impact of machine readership is more pronounced in situations where machine learning offers greater advantages, such as with complex financial statements or when alternative data is available. Furthermore, we observe an overall decrease in misstatements, suggesting that machine readership enhances overall quality of financial reporting, rather than prompting managers to redirect misreporting to areas beyond machine detection.
