Reinforcement Learning for Systematic FX Trading

We explore online inductive transfer learning, with a feature representation transfer from a radial basis function network formed of Gaussian mixture model hidden processing units to a direct, recurrent reinforcement learning agent. This agent is put to work in an experiment, trading the major spot market currency pairs, where we accurately account for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to the agent via a quadratic utility, who learns to target a position directly. We improve upon earlier work by learning to target a risk position in an online transfer learning context. Our agent achieves an annualised portfolio information ratio of 0.52 with a compound return of 9.3%, net of execution and funding cost, over a 7-year test set; this is despite forcing the model to trade at the close of the trading day at 5 pm EST when trading costs are statistically the most expensive.<br>

Risk-adjusted return, fund manager efficacy, and fund characteristics – An inquiry

In this study, I examine the risk-adjusted return of mutual funds in India. A data set of 4220 mutual funds is used for the analysis. Sharpe ratio, a metric of risk-adjusted return (Sharpe, 1994) and Information ratio, a metric of outperformance than a fund’s benchmark (Goodwin, 1998) were analyzed. Regression analysis is used to estimate the impact of fund characteristics like fund category, fund type, fund access type, corpus size on the dependent variables i.e., Sharpe Ratio and the Information Ratio. All the funds underperformed in both the Sharpe ratio and Information ratio. Liquid funds found worst. Fund type and corpus size do not impact fund performance. Fund access type was found to be significant on fund performance. The results add to the literature by examining the post-pandemic period.

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Reinforcement Learning for Systematic FX Trading

We conduct a detailed experiment on major cash fx pairs, accurately accounting for transaction and funding costs. These sources of profit and loss, including the price trends that occur in the currency markets, are made available to our recurrent reinforcement learner via a quadratic utility, which learns to target a position directly. We improve upon earlier work, by casting the problem of learning to target a risk position, in an online learning context. This online learning occurs sequentially in time, but also in the form of transfer learning. We transfer the output of radial basis function hidden processing units, whose means, covariances and overall size are determined by Gaussian mixture models, to the recurrent reinforcement learner and baseline momentum trader. Thus the intrinsic nature of the feature space is learnt and made available to the upstream models. The recurrent reinforcement learning trader achieves an annualised portfolio information ratio of 0.52 with compound return of 9.3\%, net of execution and funding cost, over a 7 year test set. This is despite forcing the model to trade at the close of the trading day 5pm EST, when trading costs are statistically the most expensive. These results are comparable with the momentum baseline trader, reflecting the low interest differential environment since the the 2008 financial crisis, and very obvious currency trends since then. The recurrent reinforcement learner does nevertheless maintain an important advantage, in that the model's weights can be adapted to reflect the different sources of profit and loss variation. This is demonstrated visually by a USDRUB trading agent, who learns to target different positions, that reflect trading in the absence or presence of cost.<br>

Investment returns from reputation investing: do good firms provide good returns?

PurposeThis study aims to examine whether a good corporate reputation leads to superior investment returns. Theory and empirics provide support for the idea that a good corporate reputation improves firm value, but much of the previous research fails to consider the risk of the companies they study and relies only on accounting measures of performance such as return on assets. A complete picture of the relationship between corporate reputation and shareholder value should include risk-adjusted returns and correlation with benchmark returns.Design/methodology/approachThe Harris Poll Reputation Quotient (RQ), based on the reputations of the 100 most visible companies, suggests that companies with a “solid reputation” are more likely to be attractive investments. The authors construct portfolios using deciles and the RQ categories, rebalancing annually as RQ rankings are updated. Returns are adjusted for risk using Jensen's alpha, the information ratio, the Sharpe ratio, Modigliani and Modigliani's M2 measure, and Muralidhar's M3 measure.FindingsThe results indicate that choosing a portfolio based on the highest RQ-ranked firms does outperform the market on a risk-adjusted basis, and that the relationship between rankings and time-weighted returns is roughly monotonic. The authors also observe that corporate reputation is persistent, and that the best and worst most-visible firms are more likely to be privately held.Originality/valueThis research adds to the literature by including both market-based return measures and risk in the examination of the relationship between corporate reputation and financial performance.

On being Sharīʿah compliant in equity investments: impact of investment horizon andmarket volatility

Purpose This paper aims to undertake a comprehensive comparative analysis of Sharīʿah-compliant equity investments (SCEIs) and their non-Sharīʿah counterparts, in India, conditioning for investment horizon and market volatility. Indirectly, it also investigates for time varying performance of SCEIs, and explicitly analyses the unsystematic risk and related adequacy of returns. Design/methodology/approach Testing for statistical significance of differences in risks and returns; analysing portfolio performance using conventional metrics, information ratio, and Jensen's Alpha; Estimating returns due to stock selection and market timing using Fama’s Net Selectivity and Treynor and Mazuy’s Models. Findings SCEIs in India do not significantly differ in their total risks and returns compared to their conventional counterparts. While their risk is lower in the monthly and quarterly investment horizons, their Jensen’s Alphas are positive only in the annual investment horizons. These findings hold, when market volatility is low. Market timing wipes out the superior returns that exist due to stock selection in SCEIs. Research limitations/implications Being Sharīʿah-compliant is beneficial only in longer investment horizons. Asset selection, not co-movement with the market, is key to excess returns to compensate for risks due to inadequate diversification. However, only cautious market timing can conserve them. Practical implications Though investors are not better-off in choosing ethical investments, they are not worse-off either. Being Sharīʿah-compliant is rewarding during less volatile markets. Originality/value This paper extends international literature on SCEIs, with evidence on the impact of investment horizon and market volatility on their returns and risks. Further, this paper is also a comprehensive analysis of Indian SCEIs, broadening the empirical evidence on a significant, non-Islamic and emerging market.

Rx TimeMachine: A global pseudoprospective earthquake forecast database for training and ranking predictive algorithms

&lt;p&gt;Recent advances in machine learning and pattern recognition methods have propagated into various applications in seismology. Phase picking, earthquake location, anomaly detection and classification applications have benefited also from the increased availability of cloud computing and open-source software libraries. However, applications of these new techniques to the problems of earthquake forecasting and prediction have remained relatively stagnant.&lt;/p&gt;&lt;p&gt;The main challenges in this regard have been the testing and validation of the proposed methods. While there are established metrics to quantify the performance of algorithms in common pattern recognition and classification problems, the earthquake prediction problem requires a properly defined reference (null) model to establish the information gain of a proposed algorithm. This complicates the development of new methods, as researchers are required to develop not only a novel algorithm but also a sufficiently robust null model to test it against.&lt;/p&gt;&lt;p&gt;We propose a solution to this problem. We have recently introduced a global real-time earthquake forecasting model that can provide occurrence probabilities for a user defined time-space-magnitude window anywhere on the globe (Nandan et al. 2020). In addition, we have proposed the Information Ratio (IR) metric that can rank algorithms producing alarm based deterministic predictions as well as those producing probabilistic forecasts (Kamer et al. 2020). To provide the community with a retrospective benchmark, we have run our model in a pseudoprospective fashion for the last 30 years (1990-2020). We have calculated and stored the earthquake occurrence probabilities for each day, for the whole globe (at ~40km resolution) for various time-space windows (7 to 30 days, 75 to 300 km). These can be queried programmatically via an Application Programmable Interface (API) allowing model developers to train and test their algorithms retrospectively. Here we shall present how the Rx TimeMachine API is used for the training of a simple pattern recognition algorithm and show the algorithm's prospective predictive performance.&lt;br&gt;&lt;br&gt;Nandan, S., Kamer, Y., Ouillon, G., Hiemer, S., Sornette, D. (2020). &lt;em&gt;Global models for short-term earthquake forecasting and predictive skill assessment&lt;/em&gt;. European Physical Journal ST. doi: 10.1140/epjst/e2020-000259-3&amp;#160;&lt;br&gt;Kamer, Y., Nandan, S., Ouillon, G., Hiemer, S., Sornette, D. (2020). &lt;em&gt;Democratizing earthquake predictability research: introducing the RichterX platform&lt;/em&gt;. European Physical Journal ST. doi: 10.1140/epjst/e2020-000260-2&amp;#160;&lt;/p&gt;