Selected Works from Bloomberg-Columbia Machine Learning in Finance Workshop（彭博社/哥伦比亚大学金融机器学习研讨会论文选读）

一年一度的Bloomberg-Columbia Machine Learning in Finance Workshop 目前已经举办了7届，作为专门针对量化金融领域里机器学习的最前沿应用方法的会议，在过去几年的活动中都涌现出了很多非常新颖和具有引导性的研究。这篇文章将按时间顺序收集一些在这个会议系列中出现的相关学术论文，包括深度学习，强化学习，生成模型，风险模型，交易执行等：

2021

Black-Box Model Risk in Finance by Samuel N. Cohen, Derek Snow, and Lukasz Szpruch

Machine learning models are increasingly used in a wide variety of financial settings. The difficulty of understanding the inner workings of these systems, combined with their wide applicability, has the potential to lead to significant new risks for users; these risks need to be understood and quantified. This paper focuses on a well studied application of machine learning techniques, to pricing and hedging of financial options. The aim will be to highlight the various sources of risk that the introduction of machine learning emphasises or de-emphasises, and the possible risk mitigation and management strategies that are available.

Synthetic Data: A New Regulatory Tool by Jay Cao, Jacky Chen, John Hull, Zissis Poulos, and Dorothy Zhang

Machine learning tools have been developed to generate synthetic data sets that are indistinguishable from available historical data. This paper investigates whether the tools can be used for stress testing. In particular they test whether synthetic data can be used to provide reliable risk measures when the confidence levels are high. The results are encouraging and suggest that synthetic data produced from the most recent 250 days of historical data are potentially useful for determining regulatory market risk capital requirements.

Deep Learning Modeling of Limit Order Book: A Comparative Perspective by Antonio Briola, Jeremy Turiel, and Tomaso Aste

This work addresses theoretical and practical questions in the domain of Deep Learning for High Frequency Trading. State-of-the-art models such as Random models, Logistic Regressions, LSTMs, LSTMs equipped with an Attention mask, CNN-LSTMs and MLPs are reviewed and compared on the same tasks, feature space, and dataset and clustered according to pairwise similarity and performance metrics. The underlying dimensions of the modelling techniques are hence investigated to understand whether these are intrinsic to the Limit Order Book's dynamics. They observe that the Multilayer Perceptron performs comparably to or better than state-of-the-art CNN-LSTM architectures indicating that dynamic spatial and temporal dimensions are a good approximation of the LOB's dynamics, but not necessarily the true underlying dimensions.

Explainable AI in Credit Risk Management by Branka Hadji Misheva, Joerg Osterrieder, Ali Hirsa, Onkar Kulkarni, and Stephen Fung Lin

Artificial Intelligence (AI) has created the single biggest technology revolution the world has ever seen. For the finance sector, it provides great opportunities to enhance customer experience, democratize financial services, ensure consumer protection and significantly improve risk management. While it is easier than ever to run state-of-the-art machine learning models, designing and implementing systems that support real-world finance applications have been challenging. In large part this is due to the lack of transparency and explainability which in turn represent important factors in establishing reliable technology. This paper implements different advanced post-hoc model agnostic explainability techniques to machine learning (ML)-based credit scoring models applied to loan performance data.

Why and How Systematic Strategies Decay by Antoine Falck, Adam Rej, and David Thesmar

Systematic strategies are known to decay out of sample. Two competing explanations have been proposed: arbitrage and overfitting. In order to pin down, which of the two forces is more relevant, this paper reproduces a large number of stock anomalies proposed in the academic literature and set out to determine characteristics that explain their decay out of sample. They use the cross-section of stock anomalies to test variables that proxy various aspects of arbitrage and overfitting. The study suggests that while some arbitrage-related variables are statistically significant, it is the overfitting variables that explain larger part of the cross-sectional variance of Sharpe decay across strategies.

Generative Adversarial Networks in Finance: An Overview by Florian Eckerli and Joerg Osterrieder

Modelling in finance is a challenging task: the data often has complex statistical properties and its inner workings are largely unknown. Deep learning algorithms are making progress in the field of data-driven modelling, but the lack of sufficient data to train these models is currently holding back several new applications. Generative Adversarial Networks (GANs) are a neural network architecture family that has achieved good results in image generation and is being successfully applied to generate time series and other types of financial data. The purpose of this study is to present an overview of how these GANs work, their capabilities and limitations in the current state of research with financial data and present some practical applications in the industry. As a proof of concept, three known GAN architectures were tested on financial time series, and the generated data was evaluated on its statistical properties, yielding solid results.

2020

Detecting and Correcting for Label Shift with Black Box Predictors by Zachary C. Lipton, Yu-Xiang Wang, and Alexander J. Smola

Faced with distribution shift between training and test set, we wish to detect and quantify the shift, and to correct our classifiers without test set labels. Motivated by medical diagnosis, where diseases (targets), cause symptoms (observations), this paper focuses on label shift, where the label marginal p(y) changes but the conditional p(x|y) does not. They propose Black Box Shift Estimation (BBSE) to estimate the test distribution p(y). BBSE exploits arbitrary black box predictors to reduce dimensionality prior to shift correction. While better predictors give tighter estimates, BBSE works even when predictors are biased, inaccurate, or uncalibrated, so long as their confusion matrices are invertible.

Forecasting Firm Material Events from 8-K Reports by Shuang Zhai and Zhujun Zhang

Based on a unique sequence-to-sequence formulation, this paper proposes a hybrid Transformer model to forecast firm material event sequences, by utilizing the firm's SEC 8-K current reports and financial ratios. The proposed model demonstrates superior prediction performance compared against traditional sequence-to-sequence models and task-specific Markov Chain Monte Carlo simulations.

Agency MBS Prepayment Model Using Neural Networks by Jiawei Zhang, Xiaojian Zhao, Joy Zhang, Fei Teng, Siyu Lin and Hongyuan Li

Artificial intelligence can reduce model fitting times from months to hours, significantly improving modeling efficiency and enabling true model back-testing and timelier understanding of prepayment trends. This AI prepayment model has demonstrated higher model accuracy and agility than our traditional model. It overcame high-dimensionality and high-nonlinearity issues associated with prepayment modeling. The AI prepayment model was able to detect new and often subtle prepayment signals that eluded traditional modeling approaches.

Optimal, Truthful, and Private Securities Lending by Emily Diana, Michael Kearns, Seth Neel, and Aaron Roth

This paper considers a fundamental dynamic allocation problem motivated by the problem of securities lending in financial markets, the mechanism underlying the short selling of stocks. A lender would like to distribute a finite number of identical copies of some scarce resource to n clients, each of whom has a private demand that is unknown to the lender. The lender would like to maximize the usage of the resource but is constrained to sell the resource at a pre-specified price per unit, and thus cannot use prices to incentivize truthful reporting. They first show that the Bayesian optimal algorithm for the one-shot problem actually incentivizes truthful reporting as a dominant strategy. Because true demands in the securities lending problem are often sensitive information that the client would like to hide from competitors, they then consider the problem under the additional desideratum of (joint) differential privacy.

Quant GANs: Deep Generation of Financial Time Series by Magnus Wiese, Robert Knobloch, and Ralf Korn

Modelling financial time series by using stochastic processes is a challenging task and a central area of research in mathematical finance. This paper explores how Quant GANs, a data-driven generative model based on generative adversarial networks (GANs), can be used as an alternative. The proposed Quant GAN consist out of a discriminator-generator pair that utilises temporal convolutional networks (TCNs). This architecture choice offers several advantages such as parallelism, stable gradients, guaranteed stationarity of the generated paths and enables the Quant GAN to capture volatility clusters and leverage effects.

2019

Significance Tests for Neural Networks by Enguerrand Horel and Kay Giesecke

Neural networks underpin many of the best-performing AI systems. Their success is largely due to their strong approximation properties, superior predictive performance, and scalability. However, a major caveat is explainability: neural networks are often perceived as black boxes that permit little insight into how predictions are being made. This paper tackles this issue by developing a pivotal test to assess the statistical significance of the feature variables of a neural network. They propose a gradient-based test statistic and study its asymptotics using nonparametric techniques. The limiting distribution is given by a mixture of chi-square distributions. The tests enable one to discern the impact of individual variables on the prediction of a neural network. The test statistic can be used to rank variables according to their influence.

Learning Quadratic Games on Networks by Yan Leng, Xiaowen Dong, Junfeng Wu, and Alex Pentland

Individuals, or organizations, cooperate with or compete against one another in a wide range of practical situations. Such strategic interactions are often modeled as games played on networks, where an individual's payoff depends not only on her action but also on that of her neighbors. The current literature has largely focused on analyzing the characteristics of network games in the scenario where the structure of the network, which is represented by a graph, is known beforehand. It is often the case, however, that the actions of the players are readily observable while the underlying interaction network remains hidden. This paper proposes two novel frameworks for learning, from the observations on individual actions, network games with linear-quadratic payoffs, and in particular the structure of the interaction network.

How to Play Fantasy Sports Strategically (and Win) by Martin Haugh and Raghav Singal

Daily Fantasy Sports (DFS) is a multi-billion dollar industry with millions of annual users and widespread appeal among sports fans across a broad range of popular sports. Building on the recent work of Hunter, Vielma and Zaman (2016), this paper provides a coherent framework for constructing DFS portfolios where they explicitly model the behavior of other DFS players. They formulate an optimization problem that accurately describes the DFS problem for a risk-neutral decision-maker in both double-up and top-heavy payoff settings. This formulation maximizes the expected reward subject to feasibility constraints and we relate this formulation to the literature on mean-variance optimization and the out-performance of stochastic benchmarks. Using this connection, they show how the problem can be reduced to the problem of solving a series of binary quadratic programs.

Universal Features of Price Formation in Financial Markets: Perspectives From Deep Learning by Justin Sirignano and Rama Cont

Using a large-scale Deep Learning approach applied to a high-frequency database containing billions of electronic market quotes and transactions for US equities, this paper uncovers nonparametric evidence for the existence of a universal and stationary price formation mechanism relating the dynamics of supply and demand for a stock, as revealed through the order book, to subsequent variations in its market price. They assess the model by testing its out-of-sample predictions for the direction of price moves given the history of price and order flow, across a wide range of stocks and time periods. The universal price formation model is shown to exhibit a remarkably stable out-of-sample prediction accuracy across time, for a wide range of stocks from different sectors.

2018

QLBS: Q-Learner in the Black-Scholes(-Merton) Worlds by Igor Halperin

This paper presents a discrete-time option pricing model that is rooted in Reinforcement Learning (RL), and more specifically in the famous Q-Learning method of RL. They construct a risk adjusted Markov Decision Process for a discrete-time version of the classical Black-ScholesMerton (BSM) model, where the option price is an optimal Q-function, while the optimal hedge is a second argument of this optimal Q-function, so that both the price and hedge are parts of the same formula. Pricing is done by learning to dynamically optimize risk-adjusted returns for an option replicating portfolio, as in the Markowitz portfolio theory. Using Q-Learning and related methods, once created in a parametric setting, the model is able to go model-free and learn to price and hedge an option directly from data, and without an explicit model of the world.

Classification-based Financial Markets Prediction using Deep Neural Networks by Matthew Dixon, Diego Klabjan, and Jin Hoon Bang

Market movement direction predictions should take many financial instruments simultaneously into account due to their correlations. This intrinsic complexity leads to thousands of possible features and thus it is appropriate for deep neural networks. This paper applies a feed forward network on thousands of features and compares it with more advanced recurrent neural networks that combine convolutional layers for feature embedding within long-short-term-memory cells.

High Frequency Market Making with Machine Learning by Matthew Dixon

This paper introduces a trade execution model to evaluate the economic impact of classifiers through backtesting. Extending the concept of a confusion matrix, it presents a 'trade information matrix' to attribute the expected profit and loss of tick level predictive classifiers under execution constraints, such as fill probabilities and position dependent trade rules, to correct and incorrect predictions. This approach directly evaluates the performance sensitivity of a market making strategy to classifier error and augments traditional market simulation based testing.

Big Data's Dirty Secret by Harvey Stein and Yan Zhang

Amidst the avalanche of articles on big data and machine learning, the phrase "after cleaning the data" is often found. This paper focuses on the work hidden behind this phrase. They analyze the types of dirty data found in financial time series, the problems caused by dirty data, and the performance of data cleaning algorithms. They extend the MSSA hole filling algorithm of Kondrashov and Ghil to improve its performance on CDS spread data, and combine it with clustering techniques from data science to detect bad data.

2017

Pursuit-Evasion Without Regret, with an Application to Trading by Lili Dworkin, Michael Kearns, and Yuriy Nevmyvaka

This paper proposes a state-based variant of the classical online learning problem of tracking the best expert. In their setting, the actions of the algorithm and experts correspond to local moves through a continuous and bounded state space. At each step, nature chooses payoffs as a function of each player's current position and action. This model therefore integrates the problem of prediction with expert advice with the stateful formalisms of reinforcement learning. Traditional no-regret learning approaches no longer apply, but they propose a simple algorithm that provably achieves no-regret when the state space is any convex Euclidean region. This paper describes a natural quantitative trading application in which the convex region captures inventory risk constraints, and local moves limit market impact.

How News and Its Context Drive Risk and Returns Around the World by Charles W. Calomiris and Harry Mamaysky

This paper develops a novel methodology for classifying the context of news articles to predict risk and return in stock markets. For a set of 52 developed and emerging market economies, they show that a parsimonious summary of news, including context-specific sentiment, predicts future countrylevel market outcomes, as measured by returns, volatilities, or drawdowns. The effect of present news on future market outcomes differs by news category, as well as across emerging and developed markets. Importantly, news stories about emerging markets contain more incremental information – in relation to known predictors of future returns – than do news stories about developed markets.

Estimating Latent Asset-Pricing Factors by Martin Lettau and Markus Pelger

This paper develops an estimator for latent factors in a large-dimensional panel of financial data that can explain expected excess returns. Statistical factor analysis based on Principal Component Analysis (PCA) has problems identifying factors with a small variance that are important for asset pricing. Their estimator searches for factors with a high Sharpe-ratio that can explain both the expected return and covariance structure. They derive the statistical properties of the new estimator based on new results from random matrix theory and show that our estimator can find asset-pricing factors, which cannot be detected with PCA, even if a large amount of data is available.

（不定期更新中。。。）