No Arbitrage and Arbitrage Pricing: A New Approach

We argue that arbitrage-pricing theories (APT) imply the existence of a low-dimensional nonnegative nonlinear pricing kernel. In contrast to standard constructs of the APT, we do not assume a linear factor structure on the payoffs. This allows us to price both primitive and derivative securities. Semi-nonparametric techniques are used to estimate the pricing kernel and test the theory. Empirical results using size-based portfolio returns and yields on bonds reject the nested capital asset-pricing model and linear APT and support the nonlinear APT. Diagnostics show that the nonlinear model is more capable of explaining variations in small firm returns.     

Do jumps contribute to the dynamics of the equity premium?

This paper investigates whether risks associated with time-varying arrival of jumps and their effect on the dynamics of higher moments of returns are priced in the conditional mean of daily market excess returns. We find that jumps and jump dynamics are significantly related to the market equity premium. The results from our time-series approach reinforce the importance of the skewness premium found in cross-sectional studies using lower-frequency data; and offer a potential resolution to sometimes conflicting results on the intertemporal risk-return relationship. We use a general utility specification, consistent with our pricing kernel, to evaluate the relative value of alternative risk premium models in an out-of-sample portfolio performance application.     

Detecting Accounting Fraud in Publicly Traded U.S. Firms Using a Machine Learning Approach

We develop a state-of-the-art fraud prediction model using a machine learning approach. We demonstrate the value of combining domain knowledge and machine learning methods in model building. We select our model input based on existing accounting theories, but we differ from prior accounting research by using raw accounting numbers rather than financial ratios. We employ one of the most powerful machine learning methods, ensemble learning, rather than the commonly used method of logistic regression. To assess the performance of fraud prediction models, we introduce a new performance evaluation metric commonly used in ranking problems that is more appropriate for the fraud prediction task. Starting with an identical set of theory-motivated raw accounting numbers, we show that our new fraud prediction model outperforms two benchmark models by a large margin: the Dechow et al. logistic regression model based on financial ratios, and the Cecchini et al. support-vector-machine model with a financial kernel that maps raw accounting numbers into a broader set of ratios.     

Exploiting social media with higher-order Factorization Machines: statistical arbitrage on high-frequency data of the S&P 500

Over the past 15 years, there have been a number of studies using text mining for predicting stock market data. Two recent publications employed support vector machines and second-order Factorization Machines, respectively, to this end. However, these approaches either completely neglect interactions between the features extracted from the text, or they only account for second-order interactions. In this paper, we apply higher-order Factorization Machines, for which efficient training algorithms have only been available since 2016. As Factorization Machines require hyperparameters to be specified, we also introduce a novel adaptive-order algorithm for automatically determining them. Our study is the first one to make use of social media data for predicting minute-by-minute stock returns, namely the ones of the S&P 500 stock constituents. We show that, unlike a trading strategy employing support vector machines, Factorization-Machine-based strategies attain positive returns after transactions costs for the years 2014 and 2015. Especially the approach applying the adaptive-order algorithm outperforms classical approaches with respect to a multitude of criteria, and it features very favorable characteristics.     

The dynamic interaction between volatility and returns in the US stock market using leveraged bootstrap simulations

One of the most important stylized facts in finance is that stock index returns are inversely related to volatility. The theoretical rationale behind the proposition is still controversial. The causal relationship between returns and volatility is investigated in the US stock market over the period 2004-2009 using daily data. We apply a bootstrap test with leveraged adjustments that is robust to non-normality and ARCH. We find that the volatility causes returns negatively and returns cause volatility positively. The policy implications of our findings are discussed in the main text.     

Trading profitability from learning and adaptation on the Tokyo Stock Exchange

This study proposes unexamined technical trading rules, which are dynamically switching strategies among filter, moving average and trading-range breakout rules. The dynamically switching strategy is formulated based on a discrete choice theory consistent with the concept of myopic utility maximization. We utilize the transaction data of the individual stocks listed on the Nikkei 225 from September 1, 2005 to August 31, 2007. We demonstrate that switching strategies produce positive returns and their performance is better than those from the buy-and-hold and non-switching strategies over our sample periods. We also demonstrate equivalent performance for switching with different learning horizons, implying that behavioural heterogeneity of stock investors arises from the coexistence of different strategies with varying degrees of learning horizons. Our result supports several research assumptions and results on agent-based theoretical models that successfully replicate empirical features in financial markets, such as fat tails of return distributions and volatility clustering. However, upon considering the effects of data-snooping bias superior performance disappears.     

Enhancing the momentum strategy through deep regression

Momentum is a pervasive and persistent phenomenon in financial economics that has been found to generate abnormal returns not explainable by the traditional asset pricing models. This paper investigates some variations of the existing momentum strategies to increase profit and gain other desirable properties such as low kurtosis, small negative skewness and small maximum drawdown. We investigate these by using regression that is based on the latest techniques from deep learning such as stacked autoencoders and denoising autoencoders. Empirical results indicate that our regression-based variations can generate increased returns, and improved higher-order moments and maximum drawdown characteristics. Furthermore, our results reveal such improved performance can only be attained through the use of the latest deep learning technologies.     

Time-varying neural network for stock return prediction

We consider the problem of neural network training in a time-varying context. Machine learning algorithms have excelled in problems that do not change over time. However, problems encountered in financial markets are often time varying. We propose the online early stopping algorithm and show that a neural network trained using this algorithm can track a function changing with unknown dynamics. We compare the proposed algorithm to current approaches on predicting monthly US stock returns and show its superiority. We also show that prominent factors (such as the size and momentum effects) and industry indicators exhibit time-varying predictive power on stock returns. We find that during market distress, industry indicators experience an increase in importance at the expense of firm level features. This indicates that industries play a role in explaining stock returns during periods of heightened risk.     

The impact of hedging on the market value of equity

We examine the annual stock performance of firms that disclose the use of derivatives to hedge over the period 1995 to 1999. We find that only 21.6% of publicly traded U.S. corporations in our sample hedged with derivative instruments over this period and their use is concentrated in the larger companies. Similar to other studies we find that when derivatives are used, interest rate and currency securities are used much more frequently than commodity products. Our sample of 1308 companies that hedge outperforms other securities by 4.3% per year on average over our sample period. This result is robust to several alternative methods of estimating abnormal returns. When we segment performance by the type of hedge used, however, we find that the over-performance is due entirely to larger firms that hedge currency. We find no abnormal returns for firms hedging either interest rates or commodities. The abnormal returns in firms hedging currency is robust to alternative models that seek to control for exchange rate fluctuations and global equity returns; however, we find no significant abnormal returns to currency hedgers when using an augmented model that controls for the role of intangible assets.     

Nonparametric Inference of Value-at-Risk for Dependent Financial Returns

The article considers nonparametric estimation of value-at-risk(VaR) and associated standard error estimation for dependentfinancial returns. Theoretical properties of the kernel VaRestimator are investigated in the context of dependence. Thepresence of dependence affects the variance of the VaR estimatesand has to be taken into consideration in order to obtain adequateassessment of their variation. An estimation procedure of thestandard errors is proposed based on kernel estimation of thespectral density of a derived series. The performance of theVaR estimators and the proposed standard error estimation procedureare evaluated by theoretical investigation, simulation of commonlyused models for financial returns, and empirical studies onreal financial return series.     

Portfolio performance measurement using APM-free kernel models

A general asset-pricing framework is used to derive a conditional asset-pricing kernel that accounts efficiently for time variation in expected returns and risk, and is suitable to perform (un)conditional evaluations of passive and dynamic investment strategies. The positive abnormal unconditional performance of Canadian equity mutual funds over the period 1989–1999 becomes negative with conditioning, and is robust to the removal of ex post index mimickers. The reversal in the size-based performance results with limited information conditioning is alleviated somewhat with an expansion of the conditioning set. The performance statistics are weakly sensitive to changes in the level of relative risk aversion of the uninformed investor. Unconditional positive performances based on averages of individual fund performances lose their significance when cross-correlations are accounted for using the block-bootstrap method. Estimates of survivorship bias due to the elimination of funds with shorter lives, which range from 36 to 58 basis points per year, are stable across performance models but differ across groupings by fund objective.     

Performance and performance persistence of UK closed-end equity funds

Using a comprehensive data set of almost 300 UK closed-end equity funds over the period 1990 to 2013, we use the false discovery rate to assess the alpha-performance of individual funds with both domestic and other mandates, using self-declared benchmarks and additional risk factors. We find evidence to indicate that up to 16% of the funds have truly positive alphas while around 3% have truly negative alphas. Positive post-formation alphas using fund-price returns depend on the factor model used: there is some positive-alpha performance when post-formation returns are evaluated using a one-factor global model but substantial positive-alpha performance when using a four-factor global model.     

Stock investment strategy combining earnings power index and machine learning

We propose an intermediate-term stock investment strategy based on fundamental analysis and machine learning. The approach uses predictors from the Earnings Power Index (EPI) as input variables derived from cross-sectional and time-series data from a company’s financial statements. The analytical methods of machine learning allow us to validate the link between financial factors and excess returns directly. We then select stocks for which returns are likely to increase at the time of the next disclosed financial statement. To verify the proposed approach’s usefulness, we use company data listed publicly on the Korean stock market from 2013 to 2019. We examine the profitability of trading strategy based on ten machine-learning techniques by forming long, short, and hedge portfolios with three different measures. As a result, most portfolios, including EPI-related variables, present positive returns regardless of the period. Especially, the neural network of the two layers with sigmoid function presents the best performance for the period of 3 months and 6 months, respectively. Our results show that incorporating machine learning is useful for mid-term stock investment. Further research into the possible convergence of financial statement analysis and machine-learning techniques is warranted.     

Semiparametric estimation of a characteristic-based factor model of common stock returns

We introduce an alternative version of the Fama–French three-factor model of stock returns together with a new estimation methodology. We assume that the factor betas in the model are smooth nonlinear functions of observed security characteristics. We develop an estimation procedure that combines nonparametric kernel methods for constructing mimicking portfolios with parametric nonlinear regression to estimate factor returns and factor betas simultaneously. The methodology is applied to US common stocks and the empirical findings compared to those of Fama and French.     

Peer performance and stock market entry

Peer performance can influence the adoption of financial innovations and investment styles. We present evidence of this type of social influence: recent stock returns that local peers experience affect an individual's stock market entry decision, particularly in areas with better opportunities for social learning. The likelihood of entry does not decrease as returns fall below zero, consistent with people not talking about decisions that have produced inferior outcomes. Market returns, media coverage, local stocks, omitted local variables, short sales constraints, and stock purchases within households do not seem to explain these results.     

A smooth non-parametric estimation framework for safety-first portfolio optimization

In this paper, we adopt a smooth non-parametric estimation to explore the safety-first portfolio optimization problem. We obtain a non-parametric estimation calculation formula for loss (truncated) probability using the kernel estimator of the portfolio returns’ cumulative distribution function, and embed it into two types of safety-first portfolio selection models. We numerically and empirically test our non-parametric method to demonstrate its accuracy and efficiency. Cross-validation results show that our non-parametric kernel estimation method outperforms the empirical distribution method. As an empirical application, we simulate optimal portfolios and display return-risk characteristics using China National Social Security Fund strategic stocks and Shanghai Stock Exchange 50 Index components.     

Can machine learning models save capital for banks? Evidence from a Spanish credit portfolio

We study the impact of machine learning (ML) models for credit default prediction in the calculation of regulatory capital by financial institutions. We do so by using a unique and anonymized database from a major Spanish bank. We first compare the statistical performance of five models based on supervised learning like Logistic Lasso, Trees (CART), Random Forest, XGBoost and Deep Learning, with a well-known model like Logit. We measure the statistical performance through different metrics, and for different sample sizes and features available. We find that ML models outperform, even when relatively low amount of data is used. We then translate this statistical performance into economic impact by estimating the savings in capital when using an advanced ML model instead of a simpler one to compute the risk-weighted assets following the Internal Ratings Based (IRB) approach. Our benchmark results show that implementing XGBoost instead of Logistic Lasso could yield savings from 12.4% to 17% in terms of regulatory capital requirements.     

THE OVERNIGHT AND DAILY TRANSMISSION OF STOCK INDEX FUTURES PRICES BETWEEN MAJOR INTERNATIONAL MARKETS

Stock index futures prices for the world's major equity markets, Japan, the UK and the US, are used to examine the interaction of international equity markets. By using stock index futures prices, we avoid the nonsynchronous data problem inherent with opening and closing market averages. We find that the US is the dominant world market; overnight returns in Japan and the UK are greatly influenced by the US daily returns. In contrast, the Japanese market has no impact on the overnight or daily returns in the UK, while the UK daily performance has a small influence on Japanese overnight returns. Slight evidence of over-reaction at the opening of Japanese futures exists as the daily Nikkei returns are negatively related to the US returns.     

Are carry,momentum and value still there in currencies?

We show that carry, momentum and value predictability in currencies is associated with mispricing. Specifically, investment performance disappears subsequent to published evidence showing portfolio returns are not fully explained by risk. Replicating these studies, we show that the average out-of-sample Sharpe ratio decreases from +0.39 to −0.32. Cross sectional tests show that currencies no longer respond to interest rate and real exchange rate differentials. During this period currency excess returns do not exhibit autocorrelation. Our results are consistent with investors learning about mispricing from academic research.