Journal of Finance and Data Science最新文献

In this paper we utilize a Long Short-Term Memory Neural Network to learn from and improve upon traditional trading algorithms used in technical analysis. The rationale behind our study is that the network can learn market behavior and be able to predict when a given strategy is more likely to succeed. We implemented our algorithm in Python pursuing Google's TensorFlow. We show that our strategy, based on a combination of neural network prediction, and traditional technical analysis, performs better than the latter alone.

This paper demonstrates how to collect and manage free predicted earnings surprises available in the public domain. The predicted earnings surprises we collect are expected to be more accurate than the corresponding consensus estimates and other predicted earnings, but have not been studied in the academic literature until very recently. We find a number of unexpected and problematic idiosyncrasies with the source of the data and the predicted earnings surprises themselves. The data are hard to work with, perhaps by design, and contain both big and small extreme values that are unexpected given their origin. It is unclear how these observations are selected for public release. After the data science exercise of managing and merging the predicted earnings surprises with other freely available public information (specifically ticker symbols and return data), we examine the predicted earnings surprises and investigate how the predicted earnings surprises affect short-term stock prices. We find evidence of a linear association between the predicted earnings surprises and subsequent short-term returns, although the significance is driven by extreme outliers. Most importantly, we use the predicted earnings surprises to form short-term trading strategies. The most profitable trading strategy that exploits the predicted earnings surprises is a contrarian trading strategy.

The use of textual information contained in company filings with the Securities Exchange Commission (SEC), including annual reports on Form 10-K, quarterly reports on Form 10-Q, and current reports on Form 8-K, has gained the increased attention of finance and accounting researchers. In this paper we use a set of machine learning methods to predict the market response to changes in a firm's auditor as reported in public filings. We vectorize the text of 8-K filings to test whether the resulting feature matrix can explain the sign of the market response to the filing. Specifically, using classification algorithms and a sample consisting of the Item 4.01 text of 8-K documents, which provides information on changes in auditors of companies that are registered with the SEC, we predict the sign of the cumulative abnormal return (CAR) around 8-K filing dates. We report the correct classification performance and time efficiency of the classification algorithms. Our results show some improvement over the naïve classification method.

In an environment characterized by stochastic variances and correlations, we demonstrate through construction of the equilibrium index option value from constituent components, that the generalized PDE identifies the stochastic elements differentially affecting index option prices relative to prices of aggregated constituent stock options. A unified treatment of the generalized partial differential system for index and constituent stock options in Theorem 1 illustrates that nonlinear interactive terms emanating from stochastic correlation affect index option price and return essentially different from contributions to the aggregated risks of the constituent stock options. Our study contributes to the growing evidence of priced correlation risk in markets for index and constituent stock options.

Theorem 1 illustrates the pricing differential, while Proposition 1 illustrates that the pricing differential produces a quantifiable metric of the measure of the nonlinear interactive terms. The quantifiable metric is constructed from the difference between the model free implied variance of the index and a weighted aggregate of the model free implied variances of the constituent stocks. Proposition 2 identifies that index variance risk premium includes additional significant contributions from the nonlinear interactive risks not present in the aggregated returns of the constituent stocks. The nonlinear interactive risks produce a wedge between the instantaneous expected excess index and aggregated stock option returns.

This paper aims to analyze the impact of public education expenditures on GDP per capita of Tunisia and Morocco during the period 1980–2015. This study is based on the Auto-Regressive Distributive Lags (ARDL) approach that is proposed by Pesaran et al. The empirical estimate yields interesting results. In the short term, the relationship between public spending on education and GDP per capita in Morocco is positive while it is negative in Tunisia. In the long term, by contrast, public expenditure on education serves to increase the GDP per capita of the two countries, but more intensively so in Morocco than in Tunisia.

I have extended the arithmetic and logarithmic equations of the daily return by including daily dividend. To do this, firstly, I have mathematically broadened the scope of the two mostly used formulas of daily return by including daily dividend. Next, I have developed a couple of daily dividend estimation models from both pre and post stockholders' perspective. While developing those models, I have functionally used the compounding factors of time value theory. Finally, I have empirically examined the statistical robustness of Model-1. The findings of the study revealed that inclusion of daily dividend significantly increased the daily and monthly arithmetic and logarithmic returns of the securities. However, after inclusion of daily dividend, the long run variances of the both arithmetic return series remains same whereas the long run variances of both logarithmic return series significantly turns down to around zero percent direct a sharp decline of the risk of logarithmic return. Moreover, after inclusion of daily dividend the Value at Risk (VaR) of the daily logarithmic return declines sharply validates Model 1 for computing the daily logarithmic return.

Return smoothing and performance persistence are both sources of autocorrelation in hedge fund returns. The practice of pre-processing the data in order to remove smoothing before conducting performance analysis also affects the predictability of hedge fund returns. This paper develops a Bayesian framework for the performance evaluation of hedge funds that simultaneously accounts for smoothing, time-varying performance and factor loadings, and the short-lived nature of reported returns. Simulation evidence reveals that “unsmoothing” predictable, persistent hedge fund returns reduces the ability to detect performance persistence in the second step of the analysis. Empirically, smoothing generates severe biases in standard estimates of abnormal performance, factor loadings, and idiosyncratic volatility. In particular, for funds with high systematic risk, a standard deviation increase in smoothing implies an upward bias in α in excess of 2% annually and a downward bias in equity market beta of more than 20%. For funds with low systematic risk exposure, the smoothing bias is most apparent in estimates of idiosyncratic volatility.

Investors in financial markets are often at a loss when facing a huge range of products. For financial institutions also, how to recommend products to the right investors, especially those without previous investment records is problematic. In this paper, we develop and apply a personalized recommendation system for the equity funds market, based on the idea of transfer learning. First, using modern portfolio theory, a profile of equity funds and investors is created. Then, the profile of investors in the stock market is applied to the fund market by the idea of transfer learning. Finally, a utility-based recommendation algorithm based on prospect theory is proposed and the performance of the method is verified by testing it on actual transaction data. This study provides a reference for financial institutions to recommend products and services to the long tail customers.

Time Dependent Kernel Density Estimation (TDKDE) used in modelling time-varying phenomenon requires two input parameters known as bandwidth and discount to perform. A Maximum Likelihood Estimation (MLE) procedure is commonly used to estimate these parameters in a set of data but this method has a weakness; it may not produce stable kernel estimates. In this article, a novel estimation procedure is developed using Artificial Neural Networks which eliminates this inherent issue. Moreover, evaluating the performance of the kernel estimation in terms of the uniformity of Probability Integral Transform (PIT) shows a significant improvement using the proposed method. A real-life application of TDKDE parameter estimation on NASDQ stock returns validates the flawless performance of the new technique.

The purpose of predictive stock price systems is to provide abnormal returns for financial market operators and serve as a basis for risk management tools. Although the Efficient Market Hypothesis (EMH) states that it is not possible to anticipate market movements consistently, the use of computationally intensive systems that employ machine learning algorithms is increasingly common in the development of stock trading mechanisms. Several studies, using daily stock prices, have presented predictive system applications trained on fixed periods without considering new model updates. In this context, this study uses a machine learning technique called Support Vector Regression (SVR) to predict stock prices for large and small capitalisations and in three different markets, employing prices with both daily and up-to-the-minute frequencies. Prediction errors are measured, and the model is compared to the random walk model proposed by the EMH. The results suggest that the SVR has predictive power, especially when using a strategy of updating the model periodically. There are also indicative results of increased predictions precision during lower volatility periods.