Journal of Finance and Data Science最新文献

In this paper, we present an integrated approach to portfolio construction and optimization, leveraging high-performance computing capabilities. We first explore diverse pairings of generative model forecasts and objective functions used for portfolio optimization, which are evaluated using performance-attribution models based on least absolute shrinkage and selection operator (LASSO). We illustrate our approach using extensive simulations of crypto-currency portfolios, and we show that the portfolios constructed using the vine-copula generative model and the Sharpe-ratio objective function consistently outperform. To accommodate a wide array of investment strategies, we further investigate portfolio blending and propose a general framework for evaluating and combining investment strategies. We employ an extension of the multi-armed bandit framework and use value models and policy models to construct eclectic blended portfolios based on past performance. We consider similarity and optimality measures for value models and employ probability-matching (“blending”) and a greedy algorithm (“switching”) for policy models. The eclectic portfolios are also evaluated using LASSO models. We show that the value model utilizing cosine similarity and logit optimality consistently delivers robust superior performances. The extent of outperformance by eclectic portfolios over their benchmarks significantly surpasses that achieved by individual generative model-based portfolios over their respective benchmarks.

This paper studies portfolio optimization through improvements of ex-ante conditional covariance estimates. We use the cross-section of stock returns over a 52-year sample to analyze trading performance by implementing the machine learning algorithm of hierarchical clustering. We find that higher out-of-sample risk-adjusted returns are achieved relative to the traditional Markowitz portfolio through hierarchical clustering using a 3-month buy-and-hold, long-only strategy. Additionally, the average change in portfolio weights at each rebalancing period is significantly lower for the portfolio formed using machine learning relative to Markowitz, decreasing investor trading costs. The results are robust to various settings and subsamples.

Central bank communications are an important tool for guiding the economy and fulfilling monetary policy goals. Natural language processing (NLP) algorithms have been used to analyze central bank communications. These outdated bag-of-words methods often ignore context and cannot distinguish who these sentiments are addressing. Recent research has introduced deep-learning-based NLP algorithms, also known as large language models (LLMs), which take context into account. This study applies LLMs to central bank communications and constructs CentralBankRoBERTa, a state-of-the-art economic agent classifier that distinguishes five basic macroeconomic agents and binary sentiment classifier that identifies the emotional content of sentences in central bank communications. The absence of large-language models in the central bank communications literature may be attributed to a lack of appropriately labeled datasets. To address this gap, we introduce our model, CentralBankRoBERTa, offering an easy-to-use and standardized tool for scholars of central bank communications.

We propose a novel asset allocation model using a Markov process of states defined by clustered efficient frontier coefficients. While most research in Markov models of the market characterize regimes using return and volatility, we instead propose characterizing these states using efficient frontiers, which provide more information on the interactions of underlying assets that comprise the market. Efficient frontiers can be decomposed to their functional form, a square-root second-order polynomial defined by three coefficients, to provide a dimensionality reduction of the return vector and covariance matrix. Each month, the proposed model hierarchically clusters the monthly coefficients data up to the current month, to characterize the market states, then defines a Markov process on the sequence of states. To incorporate these states into portfolio optimization, for each state, we calculate the tangency portfolio using only return data in that state. We then take the expectation of these weights for each state, weighted by the probability of transitioning from the current state to each state. To empirically validate our proposed model, we employ three sets of assets that span the market, and show that our proposed model significantly outperforms benchmark portfolios.

This paper proposes a novel theory, coined as Topological Tail Dependence Theory, that links the mathematical theory behind Persistent Homology (PH) and the financial stock market theory. This study also proposes a novel algorithm to measure topological stock market changes as well as the incorporation of these topological changes into forecasting realized volatility (RV) models to improve their forecast performance during turbulent periods. The results of the empirical experimentation of this study provide evidence that the predictions drawn from the Topological Tail Dependence Theory are correct and indicate that the employment of PH information allows nonlinear and neural network models to better forecast RV during a turbulent period.

We analyze a multi-period model of capital gains taxation with endogenous prices. Relative to an economy without taxation, a capital gains tax tends to lower prices and increase returns. Abstracting from tax redistribution policies, we find that a taxable investor's welfare falls, a nontaxable investor's welfare rises, and, depending on the tax rate, social welfare may either rise or fall. The taxable investor's tax-timing option increases social welfare but may either increase or decrease tax revenue. Tax rebates for capital losses have little effect on welfare or tax revenue. Implications for empirical asset pricing are identified.

The emergence of the Bitcoin cryptocurrency marked a new era of illegal transactions. Cryptocurrency provides some level of anonymity allowing its users to create an unlimited number of wallets with alias addresses, which makes it challenging to identify the actual user. This is used by criminals for the purpose of making illegal transactions. At the same time, Bitcoin stores and provides information about all committed transactions, which opens up opportunities for identifying suspicious behavior patterns in this network using data mining. The problem of detecting suspicious activity in the Bitcoin network can be solved with sufficiently high accuracy using machine learning methods. The paper provides a comparative study of various machine learning methods to solve the mentioned problem: logistic regression, decision tree, random forest, gradient boosting. Selecting hyper parameters, rebalancing the dataset, and active learning are particularly important. The most important hyperparameters of the algorithms are described. Metrics show that the gradient boosting looks the most promising. In total 38 features of bitcoin addresses were identified. The top features are presented in the paper.