Backtest Results Research Articles

Empirical Research on Multi-Factor Alpha Strategy in the A-Share Market

Among many quantitative investment means, Alpha strategy, as a typical strategy, has ushered in unprecedented development opportunities. Therefore, how to obtain excess Alpha returns stably and how to select the portfolio with higher future returns from the massive stock pool has attracted more and more attention from investors. This paper uses the ranking score method to determine the effective factors in the candidate factors and constructs a multi-factor Alpha strategy stock selection model in the A-share market. This paper first uses historical data to conduct sample analysis of stocks, studies the correlation between various indicators of listed companies and Alpha returns of individual stocks, and selects candidate factors that can effectively screen stocks with high Alpha returns. After that, the strategy model was established for back testing to test the application and feasibility of Alpha strategy in China's A-share market. Four effective factors are selected according to previous studies. And use the market data from January 2022 to December 2022 to back test the alpha strategy. The back test results show that the multi-factor Alpha strategy in this paper meets the expected performance in the A-stock market and can outperform the market to obtain excess returns. This paper studies the multi-factor Alpha strategy in the A-stock market, aiming to improve the feasibility of this strategy in the Chinese stock market, hoping to provide objective and effective investment ideas for asset managers and obtain investment methods that can obtain excess returns.

Quantitative investment prediction analysis for enterprise asset management using machine learning algorithms

Quantitative investment can manage enterprise assets better to obtain higher revenues. This paper analyzed quantitative investment prediction using machine learning algorithms. First, the support vector machine (SVM) algorithm was introduced, and stock changes were predicted by the SVM algorithm. Then, the feature factors in stock data were extracted by maximum information coefficient (MIC) as the input of the SVM algorithm. Finally, the performance and backtest results of the SVM algorithm was analyzed. It was found that the SVM algorithm had a good performance, and its F1-score was 0.9884, which was better than C4.5 and random forest algorithms. In terms of backtesting, the portfolio built based on the prediction results of the SVM algorithm obtained a higher annualized return rate when the number of stocks was small; when the number of stocks was 10, the portfolio built based on the SVM algorithm had an annualized return rate of 83.67%, a smaller maximum retracement, and a higher Sharpe ratio than the other algorithms, which balanced the risk and return well. The results demonstrate the reliability of the SVM algorithm in predicting quantitative investment, which is beneficial to achieving the optimization of enterprise asset management.

Portfolio Optimization Considering Behavioral Stocks with Return Scenario Generation

This study extends the application of behavioral portfolio optimization by estimating the return of behavioral stocks (B-stocks). With the cause-and-effect relationships of the respective irrational behaviors on the stock price movements and the unique information provided by B-stocks in terms of knowing with a calculated probability when (time duration) a specific effect (e.g., positive cumulative abnormal return) after a certain trigger point (cause of the irrational behavior) is spotted, regression analysis is applied on the information in the duration to have more accurate return estimates. To fit in the framework of behavioral portfolio optimization, the scenarios used for the optimization are generated utilizing regression analysis, based on which the safety-first scenario-based mixed-integer program is applied to obtain the optimal portfolios. This study also proposes two new types of B-stocks with corresponding operational definitions for herding and ostrich-effect, along with the previously identified over-reaction, under-reaction, and disposition-effect B-stocks. Back-test results show that the portfolios are profitable and can significantly outperform a benchmark and the market.

Research on multi factor stock selection model based on LightGBM and Bayesian Optimization

With the development of the stock market, the number of individual investors has become more and more. Because of emotional and irrational factors, the risk and instability of stock market in China have been greatly increased. Therefore, it is necessary to introduce the idea of quantitative investment into the financial field. In this paper, firstly we use random forest, XGBoost and LightGBM to conduct rolling tests on multiple factors. After parameter adjustment based on Bayesian Optimization, we find that LightGBM-Bayes has the best effect. Finally, this paper uses the multi factor stock selection model based on LightGBM-Bayes to conduct a rolling back test on CSI300 Constituent Stocks, and the back test results are better than the benchmark CSI300 index.

Dependent structure and risk analysis of S&P 500 Index's continuously rising returns and continuously falling returns

PurposeThe stock market price time series can be divided into two processes: continuously rising and continuously falling. The authors can effectively prevent the stock market from crashing by accurately estimating the risk on continuously rising returns (CRR) and continuously falling returns (CFR).Design/methodology/approachThe authors add an exogenous variable into Log-autoregressive conditional duration (Log-ACD) model, and then apply our extended Log-ACD model and Archimedean copula to estimate the marginal distribution and conditional distribution of CRR and CFR. Plus, the authors analyze the conditional value at risk (CVaR) and present back-test results of the CVaR. The back-test shows that our proposed risk estimation method has a good estimation power for the risk of the CRR and CFR, especially the downside risk. In addition, the authors detect whether the dependent structure between the CRR and CFR changes using the change point test method.FindingsThe empirical results indicate that there is no change point here, suggesting that the results on the dependent structure and risk analysis mentioned above are stable. Therefore, major financial events will not affect the dependent structure here. This is consistent with the point that the CRR and CFR can be analyzed to obtain the trend of stock returns from a more macro perspective than daily stock returns scholars usually study.Practical implicationsThe risk estimation method of this paper is of great significance in understanding stock market risk and can provide corresponding valuable information for investment advisors and public policy regulators.Originality/valueThe authors defined a new stock returns, CRR and CFR, since it is difficult to analyze and predict the trend of stock returns according to daily stock returns because of the small autocorrelation among daily stock returns.

Adjusted Extreme Conditional Quantile Autoregression with Application to Risk Measurement

In this paper, we propose an extreme conditional quantile estimator. Derivation of the estimator is based on extreme quantile autoregression. A noncrossing restriction is added during estimation to avert possible quantile crossing. Consistency of the estimator is derived, and simulation results to support its validity are also presented. Using Average Root Mean Squared Error (ARMSE), we compare the performance of our estimator with the performances of two existing extreme conditional quantile estimators. Backtest results of the one-day-ahead conditional Value at Risk forecasts are also given.

Mean–Variance Optimization for Asset Allocation

The mean–variance model is widely acknowledged as the foundation of portfolio allocation because it provides a framework for analyzing the trade-off between risk and return for gaining diversification benefits. Despite the well-known shortcomings of the model, it is often the starting point for making asset allocation decisions. In this article, the authors briefly review mean–variance optimization and approaches for resolving its limitations by demonstrating backtest results on asset allocation. Feedback from asset managers is also included to explain how optimization methods are applied in practice. <b>TOPICS:</b>Statistical methods, portfolio construction, performance measurement <b>Key Findings</b> ▪ Mean–variance optimization provides a framework for analyzing risk–return trade-offs when making asset allocation decisions. ▪ Sensitivity of the mean–variance model can be addressed with robust models, and the model generates candidate asset allocations that are informative in practice. ▪ Feedback from asset managers suggests that mean–variance optimization along with simulation and various robust methods are being applied in practice for asset allocation.

Dynamic Currency Hedging with Ambiguity

This paper establishes a general relation between investor's ambiguity and non-Gaussianity of financial asset returns. Based on that relation and utilizing a flexible non-Gaussian returns model for the joint distribution of portfolio and currency returns, we develop an ambiguity-adjusted dynamic currency hedging strategy for international investors. We propose an extended filtered historical simulation that combines Monte Carlo simulation based on volatility clustering patterns with the semi-parametric non-normal return distribution from historical data. This simulation allows us to incorporate investor's ambiguity into the dynamic currency hedging strategy algorithm that can numerically optimize an arbitrary risk measure, such as volatility, value-at-risk, or expected shortfall. The out-of-sample back-test results show that, for globally diversified investors, the derived dynamic currency hedging strategy with ambiguity is stable, robust, and highly risk reductive. It outperforms the benchmarks of constant hedging as well as dynamic approaches without ambiguity in terms of lower maximum drawdown and higher Sharpe and Sortino ratios in gross terms and net of transaction costs.

Calculation of Value-at-Risk Variance-Covariance with the Approach of Simple Cash Portfolio, Factor Models and Cash Flow

One way to calculate Value-at-Risk (VaR) is the variation-covariance method. The calculation of VaR covariance assumes stock data is normally distributed. The data needed to calculate VaR by the variance-covariance method is the covariance matrix of Bank Danamon and Bank Mandiri stock data. The main topics discussed in this paper are calculating VaR covariance with a simple cash portfolio approach, factor models and cash flow. For comparison of the use of the three approaches Backtesting, the backtest results indicate that the factor model is the best method.

Practical Applications of Alice’s Adventures in Factorland: Three Blunders That Plague Factor Investing

Practical Applications Summary In Alice’s Adventures in Factorland: Three Blunders That Plague Factor Investing , which appeared in the April 2019 issue of The Journal of Portfolio Management , Rob Arnott of Research Affiliates, LLC , Campbell R. Harvey of the Fuqua School of Business at Duke University , Vitali Kalesnik of Research Affiliates Global Advisors (Europe), Limited , and Juhani Linnainmaa of the USC Marshall School of Business discuss three common mistakes in factor investing: unrealistically high expectations of factor-based returns, drawdowns that far exceed expectations, and less benefit from diversification than most observers might expect. The authors explore each of these problems and discuss how investors can address them in factor-based portfolios. First, factor returns are likely to fall short of expectations for a variety of reasons, including data mining and backtest overfitting during factor identification, and crowding following publication of a factor. Data mining (searching the data for ideas that historically work without starting with a prior economic rationale for those ideas) paired with selection bias (throwing out anything that did not work) is a near-guarantee of implausible backtest results. Second, factor performance not normally distributed, exhibiting excessive kurtosis and negative skewness. Investors should expect occasional large, even extreme, factor drawdowns. Third, diversification of a portfolio cannot always mitigate the risk of extreme drawdowns due to cross-correlation of factors. Portfolios that combine factors (whether six of the most popular or eight second-tier factors) show drawdowns that are roughly as severe—measured in standard deviations away from the mean—as the individual factors.

An investor sentiment reward-based trading system using Gaussian inverse reinforcement learning algorithm

Investor sentiment has been shown as an important factor that influences market returns, and a number of profitable trading systems have been proposed by taking advantage of investor sentiment signals. In this paper, we aim to design an investor sentiment reward-based trading system using Gaussian inverse reinforcement learning method. Our hypothesis is that while markets interact with investor’s sentiment, there exists an intrinsic mapping between investor’s sentiment and market conditions revealing future market directions. We propose an investor sentiment reward based trading system aimed at extracting only signals that generate either negative or positive market responses. Such a reward extraction mechanism is based not only on market returns but also market volatility representing a succinct and robust feature space. The back-test results show that the proposed sentiment reward-based trading system is superior to various benchmark strategies on S&P 500 index and market-based ETFs as well as few other existing news sentiment-based trading signals. Moreover, we find that sentiment reward trading system is much more effective in a volatile market, but it is sensitive to transaction costs.

Multifactor Index Construction: A Skeptical Appraisal of Bottom-Up Approaches

In this article, the authors contrast the claims of promoters of “bottom-up” approaches for constructing multi-factor equity portfolios with relevant findings in the academic literature. In particular, the authors review findings in the academic literature that raise questions on the reliability of the link between factor scores and returns, on possibilities of overstating the backtest performance of bottom-up portfolios, and on the cost of concentration resulting from the chase of factor champions. The article shows that, while bottom-up approaches are driven by a naive belief into a fine-grain deterministic link between stock-level multi-factor exposures and returns, the empirical evidence in the asset pricing literature only supports the existence of a broad-stroke relationship. Moreover, it is emphasized that bottom-up approaches are prone to over-fitting and multiple testing biases. Without any adjustments for such biases, the backtest results of bottom-up approaches may be overstated. Finally, in the process of chasing factor champions, bottom-up portfolios tend to become highly concentrated while the academic literature stresses that diversification is crucial for the successful harvesting of factor premia. The authors conclude that findings in the academic literature on these three questions give rise to a healthy dose of skepticism concerning the superiority claims of bottom-up proponents.

Hedge fund allocation: Evaluating parametric and nonparametric forecasts using alternative portfolio construction techniques

We propose a model for constructing Asian funds of hedge funds. We compare the accuracy of forecasts of hedge fund returns using an ordinary least squares (OLS) regression model, a nonparametric regression model, and a nonlinear nonparametric model. We backtest to assess these forecasts using three different portfolio construction processes: an “optimized” portfolio, an equally-weighted portfolio, and the Kelly criterion-based portfolio. We find that the Kelly criterion is a reasonable method for constructing a fund of hedge funds, producing better results than a basic optimization or an equally-weighted portfolio construction method. Our backtests also indicate that the nonparametric forecasts and the OLS forecasts produce similar performance at the hedge fund index level. At the individual fund level, our analysis indicates that the OLS forecasts produce higher directional accuracy than the nonparametric methods but the nonparametric methods produce more accurate forecasts than OLS. In backtests, the highest information ratio to predict hedge fund returns is obtained from a combination of the OLS regression with the Fung–Hsieh eight-factor variables as predictors using the Kelly criterion portfolio construction method. Similarly, the highest information ratio using forecasts generated from a combination of the nonparametric regression using the Fung–Hsieh eight-factor model variables is achieved using the Kelly criterion portfolio construction method. Simulations using risk-adjusted total returns indicate that the nonparametric regression model generates superior information ratios than the analogous backtest results using the OLS. However, the benefits of diversification plateau with portfolios of more than 20 hedge funds. These results generally hold with portfolio implementation lags up to 12months.

Trading strategies with copulas

A new approach is proposed to identify trading opportunities in the equity market by using the information contained in the bivariate dependence structure of two equities. The relationships between the equity pairs are modelled with bivariate copulas and the fitted copula structures are utilised to identify the trading opportunities. Two trading strategies are considered that take advantage of the relative mispricing between a pair of correlated stocks and involve taking a position on the stocks when they diverge from their historical relationship. The position is then reversed when the two stocks revert to their historical relationship. Only stock-pairs with relatively high correlations are considered. The dependence structures of the chosen stock-pairs very often exhibited both upper- and lower-tail dependence, which implies that copulas with the correct characteristics should be more effective than the more traditional approaches typically applied. To identify trading opportunities, the conditional copula functions are used to derive confidence intervals for the two stocks. It is shown that the number of trading opportunities is highly dependent on the confidence level and it is argued that the chosen confidence level should take the strength of the dependence between the two stocks into account. The backtest results of the pairs-trading strategy are disappointing in that even though the strategy leads to profits in most cases, the profits are largely consumed by the trading costs. The second trading strategy entails using single stock futures and it is shown to have more potential as a statistical arbitrage approach to construct a portfolio.

Credit-Informed Tactical Asset Allocation

This paper outlines a tactical asset allocation (TAA) strategy that takes signals from the credit markets and applies them to the stock market. A power model is built using the Russell 2000 equity index and the Bank of America/Merrill Lynch High Yield B index. This model is then used in a tactical asset allocation strategy to judge whether equities are expensive or cheap relative to high yield bonds. Based on back-test results from 1997 to the present, the approach provides equity-like returns while lowering portfolio volatility.

Leading indicators for Arabian Gulf oil tanker rates

AbstractIn this paper, price drivers for the Arabian Gulf oil tanker rates were derived from the Bayesian logistic regression to form a leading indicator. A universe of price drivers was filtered based on statistical criteria and speculative backtest results. Results showed that refining margins in Asia, crude production in the Arabian Gulf, the vessel utilisation rate and Brent–Dubai spreads were the most significant price drivers of TD3. A time series of these drivers indicates that Arabian Gulf production has a declining importance relative to the Brent–Dubai spreads since 2004. A vector error correction mechanism analysis of the TD1 and TD3 benchmarks indicates that TD1 returns lag behind TD3 rates.

Predictions of Optimum Forming Conditions in Press Forming using TAGUCHI Method and FEM Simulation

Developments with short-term and lower cost are strongly required for shorten products life cycle. In press forming, many trials are needed to design of mold and to decide optimum forming condition. For these problems, FEM simulation is used to predict the result of processing instead of real test. The other side, TAGUCHI method (robust design) in also used to decide optimum processing conditions with narrow dispersion. Generally, production conditions have some factors such as tolerance, order entry, date of delivery, material price and machining cost. In the actual production stage, there're many kinds of suitable production conditions, e.g. accuracy priority, total cost priority and so on. Therefore, the optimum forming condition is changed by the order of priority. In this study, spring back of warm press forming of magnesium alloy plate is intended. First, material property of magnesium plate and friction coefficient of lubricants (Teflon sheet and two types of oil) are investigated to use for FEM calculation data. Next, spring back simulations performed under various processing conditions are decided by TAGUCHI method. Average spring back and its dispersion are estimated by TAGUCHI Method with the results. Then, productivity such as accuracy, total cost, total working time and non-defective unit rate are estimated at all forming condition combination. Finally, optimum processing conditions are decided each priority. And actual spring back test results are performed to confirm the predictions. It is concluded from the result that (1) The proposed method predicted optimum processing conditions effectively in each priority and (2) The predicted results conformed to the results of the actual spring back test.

Coping characteristics and performance in fattening pigs

Pigs vary in their individual behaviour in response to stress, and this coping behaviour can be measured with a backtest. In this test, a piglet is put on its back and escape attempts are counted during 1 min. Backtest results are linked to individual performance, and regrouping at 9 weeks of age on the basis of coping characteristics can influence production parameters. In the present study, piglets were cross-fostered at 3 days of age on the basis of backtest results, and pens with only active piglets (HR), with only reactive piglets (LR) and mixed pens were formed next to original litters. At 9 weeks animals were mixed and moved. The relation between backtest results and production performance was investigated. Cross-fostering induced a change in backtest results, resulting in approximately the same mean backtest results in each group and consequently on pen level no relations were found between initial coping behaviour and performance. Individual backtest results were linked to production parameters: active animals tended to grow faster and have a leaner carcass. If pig breeding companies would use the backtest to select the most active animals that are potentially the best performers, aggressive behaviour should also be considered (e.g. as part of the breeding goal), because of the link between activity and aggressive behaviour.

The Cointegration Alpha: Enhanced Index Tracking and Long-Short Equity Market Neutral Strategies

This paper presents several applications of cointegration based trading strategies: a classic index tracking strategy, a long-short equity market neutral strategy and a number of strategies combining index tracking and long-short market neutral. As opposed to other traditional index tracking or long-short equity strategies, the portfolio optimisation is based on cointegration rather than correlation. The first strategy aims to replicate a benchmark accurately in terms of returns and volatility, while the other seeks to minimise volatility and generate steady returns under all market circumstances. The combinations of index tracking and long-short market neutral are designed as to enhance the properties of the basic strategies. To validate the applicability of the cointegration technique to asset allocation, pioneered by Lucas (1997) and Alexander (1999), and explain how and why it works, we have employed a panel data on DJIA and its constituent stocks. When applied to constructing trading strategies in the DJIA, the cointegration technique produced encouraging results. For example, between January 1995 and December 2001 the most successful self-financing statistical arbitrage strategies returned (net of transaction and repo costs) approximately 10% with roughly 2% annual volatility and negligible correlation with the market. The comprehensive set of back-test results reported is meant to offer a detailed picture of the cointegration mechanism, and to emphasise its practical implementation issues. Its key characteristics, i.e. mean reverting tracking error, enhanced weights stability and better use of the information contained in stock prices, allow a flexible design of various funded and self-financing trading strategies, from index and enhanced index tracking, to long-short market neutral and alpha transfer techniques. Further enhancement of the strategy should target first, the identification of successful stock selection rules to supplement the simple cointegration results and second, the investigation of the potential benefits of applying optimal rebalancing rules.

A note on the influence of starting position, time of testing and test order on the backtest in pigs

The backtest determines the coping behaviour of a piglet in a standardised stress situation, which might be a measure for the coping style of that animal. Backtest results are related to other parameters such as immune responses and production. In this study, the backtest was standardised and it was studied if time of testing or the order in which animals were tested influenced backtest results. One hundred and eight-four piglets from 18 sows were tested at 3, 10 and 17 days of age. Before testing, the sow was disturbed to ensure that all piglets were awake. This standardisation of the starting position before testing did not improve the correlations between backtest results in the successive tests. No relations were found between backtest results and time of testing or test order.