Annualized Sharpe Ratio Research Articles

Deep reinforcement learning applied to statistical arbitrage investment strategy on cryptomarket

In the complex and dynamic nature of financial markets where increasingly sophisticated investment strategies are required, deep reinforcement learning (DRL) has proven successful in generating real-time investment strategies, outperforming classical models. Alongside this, statistical arbitrage strategies exploit temporary market inefficiencies to generate returns. In this regard, a novel DRL-based arbitrage method has been developed.This paper presents a unified framework that combines classical statistical arbitrage theory with DRL framework to generate investment strategies. The framework addresses the challenges of identifying similar asset portfolios, extracting signals indicating temporary price deviations, and determining optimal trading rules given market conditions. The proposed methodology involves constructing arbitrage portfolios based on cointegration relationships, removing signals from price series and portfolios, and using a DRL agent to make optimal decisions within a fixed time horizon.The empirical analysis focuses on the cryptocurrency market, known for its volatility and risk. Results demonstrate that DRL agents can generate strategies with positive returns in out-of-sample periods 79.52% to 112.82% with no transaction cost, outperforming market benchmark Bitcoin 32.51% return, the best performing over the period Litecoin with 57.11% return and the worst performing Solana with a 35.70% loss. Moreover, these strategies effectively reduce risk, achieving higher risk-adjusted returns on individual assets. The strategies maintain positive returns when considering transaction costs, with the DRL agent outperforming the standard arbitrage strategy. The best-performing strategy is based on a Deep Q-Network (DQN) agent with a return of 18.39%, annualized volatility of 12.22%, and an annualized Sharpe ratio of 2.43. At the same time, Bitcoin holds an annualized volatility of 44.13% and a Sharpe ratio of 1.08.Their randomness and coherence are studied to verify the robustness of the agents’ decisions. The actions generated by the agents are not random but based on well-founded policies, which align with the obtained results.

A portfolio trading system using a novel pixel graph network for stock selection and a mean-CDaR optimization for portfolio rebalancing

Due to the complexity and dynamic nature of financial markets, portfolio management tasks require continuous adaptation with market intelligence. Furthermore, to make profitable trading decisions, it is essential to predict the future performance of the market. To that end, this study proposes a dynamic portfolio trading system in two stages: stock trend prediction and portfolio optimization. In the first stage, although many researchers proposed powerful classification models to predict BUY/SELL trading signals, some scholars only focused on converting time series' features into multichannel images without considering any spatial relationships between features. Considering the relational capacity of graphs, this study proposes a novel image-based classification model, called Pixel Graph Network (PGN), to analyze the constructed images with propagating information from all neighboring pixels, build up a representation of the images as special graphs, and get advantages of a graph-based learning algorithm. In the second stage, stocks with predicted BUY trading signals in the following days, are fed into the Mean–Conditional Drawdown at Risk (M-CDaR) optimization model to prevent any significant drawdown of the investment portfolio. Moreover, a hierarchical feature selection algorithm has been proposed, combined with a continuous trend labeling method to improve the efficiency of the training process. To demonstrate the superiority of our proposed PGN model in comparison with benchmarks, 18 stocks from the New York Stock Exchange (NYSE) are used for numerical experiments. The obtained results show the improvement and effectiveness of the proposed model in both stock trend prediction and portfolio optimization. In terms of financial performance, the proposed PGN+M-CDaR portfolio-based trading system achieves an average annual return of 112%, an average annual Sortino ratio of 10.78, and an average annual Sharpe ratio of 2.79 which are promising to achieve substantial profits while mitigating downside risk.

Oil Price Exposure and the Cross-Section of Stock Returns

Abstract We provide evidence that equity investors are slow to process information about how current oil price changes affect future earnings announcements. Stock prices respond to lagged quarterly oil price changes when firms start announcing earnings in the next quarter. A cross-sectional equity trading strategy that exploits this predictability yields an annualized Sharpe ratio of 0.50. Our oil-response forecast strategy earns especially high returns after large absolute oil price changes, in recessions or bear markets, and during peak earnings season. The predictability we document is consistent with limited attention, is not driven by risk factor exposure, and survives several robustness tests. (JEL G10, G11, G14, G40, Q41)

An overview of technical analysis in systematic trading strategies returns and a novel systematic strategy yielding positive significant returns

This paper contributes to the literature on systematic trading strategies, in particular technical analysis profitability. We measure the profitability and forecasting power of a trend following strategy implemented in Python on a wide perimeter (205 European stocks, 11 industries, 7 major stock exchanges) over 8 years: from 2015 to 2022. The strategy signal is based on 4 moving averages and a trailing stop loss. We also introduce a mechanism based on trailing upper and lower price bounds to avoid false signals and limit transaction costs during lateral movements. We calibrate the iper-parameters to all stocks belonging to the same industry. The returns of the strategy applied to the constituents of the top performing industries provides a total return of 20% net of transaction costs, with an annualized Sharpe ratio of 0.54, in the out of sample time window from 2020 to 2022.

Forecasting Korean Stock Returns with Machine Learning

AbstractThis paper aims to evaluate the predictive power of financial variables by using various machine learning methods. An analysis is conducted on data for the Korean stock market, which is representative of emerging markets, over 32 years from 1987 to 2018. The study shows that median regression is a more efficient tool than mean regression in the presence of potential heterogeneity of stocks, significantly improving performance in terms of average realized monthly return. This suggests that median regression can have better predictive performance in emerging markets where there are likely to be outliers. Additionally, a gradient boosting machine (GBM) is found to be better than a traditional linear model both in prediction accuracy and portfolio performance. The hedged return from GBM is on average 2.89% per month with an annualized Sharpe ratio of 0.93 in the median regression. The neural network (NN) is also tested and shown to perform best when the number of hidden layers is two or three. Finally, we evaluatea list of predictor variables with various measures of variable importance. Variables of risk, price trend and liquidity are found to serve as important predictors.

Asset Allocation Strategy Based on Announcements and Machine Learning-- An approach in Chinese Market

Asset allocation strategy is frequently discussed by investors, either based on fundamental or quantitative analysis. This article discusses Announcement based quantitative asset allocation strategy using machine learning models, based on 2018.Q1 to 2022.Q2 Chinese A share market. We initially select and adjust the pool of tickets through announcement signals, then use technical analysis methods with machine learning models to make return predictions. Finally, we construct portfolios using Mean-variance optimization on daily frequency. The result shows that the combination of fundamental analysis and machine learning models can generate satisfactory return. The best model can reach annualized return of 59.4% considering turnover fee, beating the market which has annualized return of 3%. The annual sharpe ratio with turnover fee of the best portfolio is 2.28, which is a satisfactory result for investors. Besides, through combining fundamental analysis with quantitative methods, the interpretability and stability of quantitative models are greatly enhanced, which provides a novel way in synthesizing two separate investment concepts. In sum, this paper can provide investors with a relatively novel investment strategy that based on the impact of announcement information on stock price and the combination of fundamental and technical analysis.

Job postings and aggregate stock returns

The job openings-to-employment ratio (JOE), defined as the number of job postings divided by the employment level, is among the strongest known predictors of the equity premium. We find that JOE outperforms a broad set of over two dozen popular predictor variables in both in-sample and out-of-sample forecasting tests. Forecasts based on JOE also produce gains of 2.91% in annualized certainty equivalent return and 0.20 in annualized Sharpe ratio relative to forecasts based on the historical mean equity premium. The empirical results are consistent with a standard production-based asset pricing model with labor inputs and search frictions.

Enhancing the profitability of lottery strategies

Recent studies show that lottery strategies, buying non-lottery type stocks and shorting lottery-type stocks, earn positive returns on average. This study examines whether the profitability of lottery strategies is predictable, and, more importantly, whether such predictability is exploited to enhance their performance. As a predictor, we employ the speculation sentiment index recently developed by Davies (forthcoming) based on observable trading activity in the leveraged Exchange Traded Funds market. We find that the profitability of lottery strategies is predictable by the lagged speculation sentiment index both in sample and out of sample. We propose active trading rules that are implementable in real time and dynamically switch the long and short legs on the lottery strategies exploiting the predictive power of speculation sentiment. The proposed dynamic strategies significantly outperform the passive strategies, yielding significant economic gains for investors with certainty equivalent return gains of 7.41%–26.35% and increases in annualized Sharpe ratios of 0.37–1.15. • The profitability of lottery strategies is predicable both in- and out-of-sample. • Such predictability is exploited to enhance the performance of lottery strategies. • The dynamic lottery strategies significantly outperform the standard strategies.

Forecasting and trading credit default swap indices using a deep learning model integrating Merton and LSTMs

Using macroeconomic and financial conditions to forecast credit default swap (CDS) spreads is a challenging task. In this paper, we propose the Merton-LSTM model, a modified LSTM model formed by integrating with the Merton determinants model, to forecast the CDS indices. We provide the rigorous math behind the Merton-LSTM model, which demonstrates that by leveraging the nonlinear learning ability of LSTM with increased model capacity, the Merton-LSTM model is expected to learn the inherent association between the Merton determinants and CDS spreads. Further, the Merton-LSTM model is compared with the machine learning models LSTM, gated recurrent unit (GRU), multilayer perceptron network (MLP), support vector machine (SVM) and a typical stochastic series model in forecasting the two most liquid five-year CDS indices, North America High Yield index (CDX.NA.HY) and North America Investment Grade index (CDX.NA.IG) through the root mean squared error (RMSE) and the Diebold-Mariano test. The comparison results show that the RMSEs of the Merton-LSTM model are the lowest (6.2570–27.2000 for CDX.NA.HY and 1.3168–6.4772 for CDX.NA.IG) compared to other competitive models. The superiority of the Merton-LSTM model in forecasting performance is highlighted in long-term prediction even with a forecasting horizon extended to 28 days. Simulated trading with different thresholds and horizons is conducted in this study. We find that the Merton-LSTM trading strategy yields the highest annualized Sharpe ratios and lowest maximum losses at most thresholds and horizons, highlighting the economic significance of the proposed model.

A Bayesian-based classification framework for financial time series trend prediction.

Financial time series have been extensively studied within the past decades; however, the advent of machine learning and deep neural networks opened new horizons to apply supercomputing techniques to extract more insights from the underlying patterns of price data. This paper presents a tri-state labeling approach to classify the underlying patterns in price data into up, down and no-action classes. The introduction of a no-action state in our novel approach alleviates the burden of denoising the dataset as a preprocessing task. The performance of our labeling algorithm is experimented with using machine learning and deep learning models. The framework is augmented by applying the Bayesian optimization technique for the selection of the best tuning values of the hyperparameters. The price trend prediction module generates the required trading signals. The results show that the average annualized Sharpe ratio as the trading performance metric is about 2.823, indicating the framework produces excellent cumulative returns.

Aggregate 52-week high, limited attention, and time-varying momentum profits

We propose an aggregate 52-week high ratio (AH52) to proxy for scarcity of investor attention and show that AH52 positively predicts future momentum profits. The momentum strategy is profitable as one standard deviation increase in AH52 raises momentum profits by 0.90% per month. AH52 subsumes the predictive power of well-documented market state variables such as market illiquidity, market volatility, and down market state. A timing strategy based on AH52 exhibits a higher annualized Sharpe ratio than that of a passive buy-and-hold strategy. The predictive power of AH52 is robust across market capitalizations, sub-periods, alternative measures of aggregate 52-week high, G7 countries, the inclusion of market-wide information, and various momentum strategies.

Smart money in China's A-share market: Evidence from big data

In this study, we propose a bar-level tracker of smart money trades by detecting the trading aggressiveness of informed traders. Accordingly, we define monthly smart money measures to identify the direction of informed trades. By using the 1-min transaction data in China’s A-share market from 1999 to 2019, we find a negative cross-sectional relationship between the smart money measures and stock future returns. We construct smart money strategies based on the fuzzy c-means (FCM) clustering algorithm, wherein the optimal strategy produces an annual Sharpe ratio of 1.169. Our findings indicate that the FCM-based portfolio formation could outperform the conventional sorting-based strategies. This study may be useful for research on smart money trades and the analysis of cross-sectional variation in returns.

Resolution of financial market uncertainty around the release of unemployment rate announcements

We provide evidence that the release of the unemployment rate announcement unconditionally leads to financial market uncertainty resolution in the stock, treasury, commodity, and foreign currency markets. The finding is economically valuable. A simple daily strategy of selling the 10-year Treasury Note Volatility Index futures before the unemployment rate announcement and closing the position after the announcement generates an annualized Sharpe ratio of 3.79, while a similar intraday strategy using VIX futures generates an annualized Sharpe ratio of 3.98. Although this resolution is not conditional of the value of the unemployment rate surprise, we also find that larger (lower) than expected unemployment can weaken (strengthen) the uncertainty resolution process.

Hedge fund portfolio selection with fund characteristics

This paper examines hedge fund portfolio selection approaches in isolation and in the context of an investor’s overall portfolio. Characteristics-based portfolios that minimize risk delivers superior out-of-sample performance. For instance, a minimum variance portfolio tilting toward small funds with high alpha and strategy distinctiveness index and low systematic risk delivers an annualized Sharpe ratio of 2.03 with a maximum drawdown of 5.20%. Investors realize diversification benefits by shifting a portion of their wealth from 60 to 40 equity-bond portfolio to characteristics-based hedge fund portfolio. Investors recognize the attractiveness of characteristics-based portfolios, but do not target flows enough to erode their superior performance.

Improving stock trading decisions based on pattern recognition using machine learning technology.

PRML, a novel candlestick pattern recognition model using machine learning methods, is proposed to improve stock trading decisions. Four popular machine learning methods and 11 different features types are applied to all possible combinations of daily patterns to start the pattern recognition schedule. Different time windows from one to ten days are used to detect the prediction effect at different periods. An investment strategy is constructed according to the identified candlestick patterns and suitable time window. We deploy PRML for the forecast of all Chinese market stocks from Jan 1, 2000 until Oct 30, 2020. Among them, the data from Jan 1, 2000 to Dec 31, 2014 is used as the training data set, and the data set from Jan 1, 2015 to Oct 30, 2020 is used to verify the forecasting effect. Empirical results show that the two-day candlestick patterns after filtering have the best prediction effect when forecasting one day ahead; these patterns obtain an average annual return, an annual Sharpe ratio, and an information ratio as high as 36.73%, 0.81, and 2.37, respectively. After screening, three-day candlestick patterns also present a beneficial effect when forecasting one day ahead in that these patterns show stable characteristics. Two other popular machine learning methods, multilayer perceptron network and long short-term memory neural networks, are applied to the pattern recognition framework to evaluate the dependency of the prediction model. A transaction cost of 0.2% is considered on the two-day patterns predicting one day ahead, thus confirming the profitability. Empirical results show that applying different machine learning methods to two-day and three-day patterns for one-day-ahead forecasts can be profitable.

Option Return Predictability

Abstract We uncover new return predictability in the cross-section of delta-hedged equity options. Expected returns to writing delta-hedged calls are negatively correlated with stock price, profit margin, and firm profitability, but positively correlated with cash holding, cash flow variance, new shares issuance, total external financing, distress risk, and dispersion of analysts’ forecasts. Our option portfolio strategies have annual Sharpe ratio above two and remain profitable after transaction costs. Their profits can be explained by two option factors, while equity risk factors have no explanatory power. We find support for several economic channels at work, yet the option return predictability remains puzzling.

Forecasting and trading cryptocurrencies with machine learning under changing market conditions

This study examines the predictability of three major cryptocurrencies—bitcoin, ethereum, and litecoin—and the profitability of trading strategies devised upon machine learning techniques (e.g., linear models, random forests, and support vector machines). The models are validated in a period characterized by unprecedented turmoil and tested in a period of bear markets, allowing the assessment of whether the predictions are good even when the market direction changes between the validation and test periods. The classification and regression methods use attributes from trading and network activity for the period from August 15, 2015 to March 03, 2019, with the test sample beginning on April 13, 2018. For the test period, five out of 18 individual models have success rates of less than 50%. The trading strategies are built on model assembling. The ensemble assuming that five models produce identical signals (Ensemble 5) achieves the best performance for ethereum and litecoin, with annualized Sharpe ratios of 80.17% and 91.35% and annualized returns (after proportional round-trip trading costs of 0.5%) of 9.62% and 5.73%, respectively. These positive results support the claim that machine learning provides robust techniques for exploring the predictability of cryptocurrencies and for devising profitable trading strategies in these markets, even under adverse market conditions.

Asset Pricing with Panel Trees Under Global Split Criteria

We introduce a class of interpretable tree-based models (P-Trees) for analyzing panel data, with iterative and global (instead of recursive and local) splitting criteria to avoid overfitting and improve model performance. We apply P-Tree to generate a stochastic discount factor model and test assets for cross-sectional asset pricing. Unlike other tree algorithms, P-Trees accommodate imbalanced panels of asset returns and grow under the no-arbitrage condition. P-Trees also graphically capture nonlinearity and interaction effects and accommodate regime-switching and interactions between macroeconomic states and firm characteristics. For example, P-Tree identifies inflation as the most important macro predictor with regime-switching in U.S. equity data. Based on multiple pricing, prediction, and investment metrics, we find that (boosted or time-series) P-Trees outperform standard factor models and PCA latent factor models. An equally-weighted portfolio for five factors generated by P-Trees delivers an excess alpha of 1.09% against the Fama-French 3-factor benchmark, producing an annualized Sharpe ratio of 1.98 out-of-sample. Data-driven cutpoints in P-Trees reveal that long-run reversal, volume volatility, and industry-adjusted market equity drive cross-sectional return variations, consistent with variable importance analysis using random forests.

A Composite Four-Factor Model in China

We investigate investors’ overreaction and underreaction and their implications to asset pricing in China stock market. The study first picks anomaly variables representing investors’ overreaction and underreaction and then measures these two effects quantitatively. Both of them deliver significant excess returns, both statistically and economically, in China stock market. We then equip these two effects with the market and the size factor to construct a composite four-factor model and study how they price other assets. Extensive empirical analysis shows that this new model is suitable for China stock market. The maximum annual Sharpe ratio spanned by the four factors is 2.02, which is one time higher than those spanned by similar models such as Stambaugh and Yuan (2017) and Daniel, Hirshleifer and Sun (2020). In addition, using 149 anomaly candidates as test assets, the composite four-factor model exhibit good pricing capability, as there is only one test asset whose abnormal return given the model exceeds the 3.0 t-statistic threshold.

VIX and liquidity premium

Previous studies find as the VIX goes up, the return and the Sharpe ratio on liquidity provision increase. We argue these two phenomena are correlated because they depend on the same fundamentals: investors’ risk aversion, asset variances and asset correlations. Our theoretical model shows (1) when investors are more risk-averse, they expect a higher return for providing liquidity, (2) when assets are volatile, liquidity shocks create stronger trading demands and thus liquidity demanders pay a higher premium, and (3) when assets are highly correlated, the higher risk of spillover of liquidity shocks across assets raises the price of liquidity. An increase in any of these three factors, besides increasing the expected return and the Sharpe ratio of liquidity providers, leads to a higher value for the VIX index. Our empirical analyses show that one standard-deviation increase in each of these three factors raise liquidity providers’ expected daily return (annualized Sharpe ratio) by 0.16%, 0.38% and 0.40% (0.82, 1.27 and 2.10 units), respectively.