Strongly Typed Genetic Programming Research Articles

Deep Heterogeneous AutoML Trend Prediction Model for Algorithmic Trading in the USD/COP Colombian FX Market Through Limit Order Book (LOB)

This study presents a novel and competitive approach for algorithmic trading in the Colombian US dollar inter-bank market (SET-FX). At the core of this strategy is an advanced predictive model, developed using the Tree-based Pipeline Optimization Tool (TPOT). TPOT, an automated machine learning platform based on strongly-typed genetic programming, incorporates the Non-dominated Sorting Genetic Algorithm II (NSGA-II). This multi-objective evolutionary algorithm is instrumental in identifying machine learning models that strike an optimal balance between high accuracy and low complexity, thereby advancing the field of predictive modeling in financial markets.

Trading support system for portfolio construction using wisdom of artificial crowds and evolutionary computation

Effective portfolio management requires vast quantities of information and accurate forecasts to make decisions that generate a profitable strategy. In this study, we propose a framework that extracts useful information from Virtual Experts, which are generated using Strongly-Typed Genetic Programming. Specifically, we created a Virtual Expert pool that provides different recommendations on selling or purchasing of a particular stock, and then applied a Wisdom of Artificial Crowds post-processing algorithm to decide which action to take. We call this framework Community of Virtual Expert Investors, and it is evaluated on different metrics of risk. Results show that this approach manages to outperform a Buy and Hold strategy in a long-term scenario, both in return and in Conditional Sharpe Ratio measures. To test the robustness of these results, a bootstrapping test was performed, in which the general findings of the results were maintained.

The efficiency of Bitcoin: A strongly typed genetic programming approach to smart electronic Bitcoin markets

Cryptocurrencies have gained a lot of attention since Bitcoin was first proposed by Satoshi Nakamoto in 2008, highlighting the potential to play a significant role in e-commerce. However, relatively little is known about cryptocurrencies, their price behaviour, how quickly they incorporate new information and their corresponding market efficiency. To extend the current literature in this area, we develop four smart electronic Bitcoin markets populated with different types of traders using a special adaptive form of the Strongly Typed Genetic Programming (STGP)-based learning algorithm. We apply the STGP technique to historical data of Bitcoin at the one-minute and five-minute frequencies to investigate the formation of Bitcoin market dynamics and market efficiency. Through a plethora of robust testing procedures, we find that both Bitcoin markets populated by high-frequency traders (HFTs) are efficient at the one-minute frequency but inefficient at the five-minute frequency. This finding supports the argument that at the one-minute frequency investors are able to incorporate new information in a fast and rationale manner and not suffer from the noise associated with the five-minute frequency. We also contribute to the e-commerce literature by demonstrating that zero-intelligence traders cannot reach market efficiency, therefore providing evidence against the hypothesis of Hayek (1945; 1968). One practical implication of this study is that we demonstrate that e-commerce practitioners can apply artificial intelligence tools such as STGP to conduct behaviour-based market profiling.

Strongly-typed genetic programming and fuzzy inference system: An embedded approach to model and generate trading rules

Generating trading signals is an interesting topic and a hard problem to solve. This work uses fuzzy inference system (FIS) and strongly typed genetic programming (STGP) to generate trading rules for the US stock market, a framework that we call FISTGP. The two embedded models have not been widely evaluated in financial applications, and according to the literature, their combination could improve forecasting performance. The fitness function used to train the STGP model is based on accuracy, optimizing the buy and sell signals, taking a different approach to the classic optimization of return–risk ratio. The rules are generated in a FIS framework, and the final signal depends on the amount of information that the investor relies on. The model is suited to each investor as a recommendation of when to change portfolio composition according to his or her particular criteria. Ternary rules are generated based on an economic interpretation, considering the risk-free rate as a part of more demanding rules. The model is applied to 90 of the most traded and active stocks in the US stock market. This approach generates important recommendations and delivers useful information to investors. The results show that the proposed model outperforms the Buy and Hold (B&H) strategy by 28.62% in the test period, considering excesses of return, with almost the same risk (1.28% higher). The other base models underperform in comparison to the B&H, with the proposed model also outperforming them.

Forecasting Financial Markets Using High-Frequency Trading Data: Examination with Strongly Typed Genetic Programming

ABSTRACTMarket regulators around the world are still debating whether high-frequency trading (HFT) plays a positive or negative role in market quality. We develop an artificial futures market populated with high-frequency traders (HFTs) and institutional traders using Strongly Typed Genetic Programming (STGP) trading algorithm. We simulate real-life futures trading at the millisecond time frame by applying STGP to E-Mini S&P 500 data stamped at the millisecond interval. A direct forecasting comparison between HFTs and institutional traders indicate the superiority of the former. We observe that the negative implications of high-frequency trading in futures markets can be mitigated by introducing a minimum resting trading period of less than 50 milliseconds. Overall, we contribute to the e-commerce literature by showing that minimum resting trading order period of less than 50 milliseconds could lead to HFTs facing a queuing risk resulting in a less harmful market quality effect. One practical implication of our study is that we demonstrate that market regulators and/or e-commerce practitioners can apply artificial intelligence tools such as STGP to conduct trading behavior-based profiling. This can be used to detect the occurrence of new HFT strategies and examine their impact on the futures market.

High‐frequency trading from an evolutionary perspective: Financial markets as adaptive systems

AbstractThe recent rapid growth of algorithmic high‐frequency trading strategies makes it a very interesting time to revisit the long‐standing debates about the efficiency of stock prices and the best way to model the actions of market participants. To evaluate the evolution of stock price predictability at the millisecond timeframe and to examine whether it is consistent with the newly formed adaptive market hypothesis, we develop three artificial stock markets using a strongly typed genetic programming (STGP) trading algorithm. We simulate real‐life trading by applying STGP to millisecond data of the three highest capitalized stocks: Apple, Exxon Mobil, and Google and observe that profit opportunities at the millisecond time frame are better modelled through an evolutionary process involving natural selection, adaptation, learning, and dynamic evolution than by using conventional analytical techniques. We use combinations of forecasting techniques as benchmarks to demonstrate that different heuristics enable artificial traders to be ecologically rational, making adaptive decisions that combine forecasting accuracy with speed.

The rise of the machines in commodities markets: new evidence obtained using Strongly Typed Genetic Programming

Market regulators around the world are still debating whether or not high-frequency trading (HFT) is beneficial or harmful to market quality. We develop artificial commodities market populated with HFT scalpers and traditional commodities traders using Strongly Typed Genetic Programming (STGP) trading algorithm. We simulate real-life commodities trading at the millisecond timeframe by applying STGP to the S&P GSCI data stamped at the millisecond interval. We observe that HFT scalpers anticipate the order flow leading to severe damages to institutional traders. To mitigate the negative implications of HFT scalpers on commodities markets, we propose a minimum resting trading order period of more than 150 ms.

Front‐Running Scalping Strategies and Market Manipulation: Why Does High‐Frequency Trading Need Stricter Regulation?

AbstractRegulators continue to debate whether high‐frequency trading (HFT) is beneficial to market quality. Using Strongly Typed Genetic Programming (STGP) trading algorithm, we develop several artificial stock markets populated with HFT scalpers and strategic informed traders. We simulate real‐life trading in the millisecond time frame by applying STGP to real‐time and historical data from Apple, Exxon Mobil, and Google. We observe that HFT scalpers front‐run the order flow, resulting in damage to market quality and long‐term investors. To mitigate these negative implications, we propose batch auctions every 30 milliseconds of trading.

A note on the relationship between high-frequency trading and latency arbitrage

We develop three artificial stock markets populated with two types of market participants — HFT scalpers and aggressive high frequency traders (HFTrs). We simulate real-life trading at the millisecond interval by applying Strongly Typed Genetic Programming (STGP) to real-time data from Cisco Systems, Intel and Microsoft. We observe that HFT scalpers are able to calculate NASDAQ NBBO (National Best Bid and Offer) at least 1.5ms ahead of the NASDAQ SIP (Security Information Processor), resulting in a large number of latency arbitrage opportunities. We also demonstrate that market efficiency is negatively affected by the latency arbitrage activity of HFT scalpers, with no countervailing benefit in volatility or any other measured variable. To improve market quality, and eliminate the socially wasteful arms race for speed, we propose batch auctions in every 70ms of trading.

Return predictability and the ‘wisdom of crowds’: Genetic Programming trading algorithms, the Marginal Trader Hypothesis and the Hayek Hypothesis

We develop profitable stock market forecasts for a number of financial instruments and portfolios using a special adaptive form of the Strongly Typed Genetic Programming (STGP)-based trading algorithm. The STGP-based trading algorithm produces one-day-ahead return forecasts for groups of artificial traders with different levels of intelligence and different group sizes. The performance of the algorithm is compared with a number of benchmark forecasts and these comparisons clearly demonstrate the short-term superiority of the STGP-based method in many circumstances. Subsequently we provide detailed analysis of the impact of trader cognitive abilities and trader numbers on the accuracy of forecasting rules which allows us to conduct new experimental tests of the Marginal Trader and the Hayek Hypotheses. We find little support for the Marginal Trader Hypothesis but some evidence for the Hayek Hypothesis.

New evidence about the profitability of small and large stocks and the role of volume obtained using Strongly Typed Genetic Programming

We employ a special adaptive form of the Strongly Typed Genetic Programming (STGP)-based learning algorithm to develop trading rules based on a survival of the fittest principle. Employing returns data for the Russell 1000, Russell 2000 and Russell 3000 indices the STGP method produces greater returns compared to random walk benchmark forecasts, and the forecasting models are statistically significant in respect of their predictive effectiveness for all three indices both in- and out-of-sample. Using one-step-ahead STGP models to investigate the differences in return patterns between small and large stocks we demonstrate the superiority of models developed for small-cap stocks over those developed for large-cap stocks, indicating that small stocks are more predictable. We also investigate the relationship between trading volume and returns, and find that trading volume has negligible predictive strength, implying it is not advantageous to develop volume-based trading strategies.

A note on the relationship between market efficiency and adaptability – New evidence from artificial stock markets

We developed various artificial stock markets populated with different numbers of traders using a special adaptive form of the Strongly Typed Genetic Programming (STGP)-based learning algorithm. We then applied the STGP technique to historical data from three indices – the FTSE 100, S&P 500, and Russell 3000 – to investigate the formation of stock market dynamics and market efficiency. We used several econometric techniques to investigate the emergent properties of the stock markets. We have found that the introduction of increased heterogeneity and greater genetic diversity leads to higher market efficiency in terms of the Efficient Market Hypothesis (EMH), demonstrating that market efficiency does not necessarily correlate with rationality assumptions. We have also found that stock market dynamics and nonlinearity are better explained by the evolutionary process associated with the Adaptive Market Hypothesis (AMH), because different trader populations behave as an efficient adaptive system evolving over time. Hence, market efficiency exists simultaneously with the need for adaptive flexibility. Our empirical results, generated by a reduced number of boundedly rational traders in six of the stock markets, for each of the three financial instruments do not support the allocational efficiency of markets, indicating the possible need for governmental or regulatory intervention in stock markets in some circumstances.

Does high frequency trading affect technical analysis and market efficiency? And if so, how?

In this paper we investigate how high frequency trading affects technical analysis and market efficiency in the foreign exchange (FX) market by using a special adaptive form of the Strongly Typed Genetic Programming (STGP)-based learning algorithm. We use this approach for real one-minute high frequency data of the most traded currency pairs worldwide: EUR/USD, USD/JPY, GBP/USD, AUD/USD, USD/CHF, and USD/CAD. The STGP performance is compared with that of parametric and non-parametric models and validated by two formal empirical tests. We perform in-sample and out-of-sample comparisons between all models on the basis of forecast performance and investment return. Furthermore, our paper shows the relative strength of these models with respect to the actual trading profit generated by their forecasts. Empirical experiments suggest that the STGP forecasting technique significantly outperforms the traditional econometric models. We find evidence that the excess returns are both statistically and economically significant, even when appropriate transaction costs are taken into account. We also find evidence that HFT has a beneficial role in the price discovery process.

Herd behaviour experimental testing in laboratory artificial stock market settings. Behavioural foundations of stylised facts of financial returns

Many scholars express concerns that herding behaviour causes excess volatility, destabilises financial markets, and increases the likelihood of systemic risk. We use a special form of the Strongly Typed Genetic Programming (STGP) technique to evolve a stock market divided into two groups—a small subset of artificial agents called ‘Best Agents’ and a main cohort of agents named ‘All Agents’. The ‘Best Agents’ perform best in term of the trailing return of a wealth moving average. We then investigate whether herding behaviour can arise when agents trade Dow Jones, General Electric, and IBM financial instruments in four different artificial stock markets. This paper uses real historical quotes of the three financial instruments to analyse the behavioural foundations of stylised facts such as leptokurtosis, non-IIDness, and volatility clustering. We found evidence of more herding in a group of stocks than in individual stocks, but the magnitude of herding does not contribute to the mispricing of assets in the long run. Our findings suggest that the price formation process caused by the collective behaviour of the entire market exhibit less herding and is more efficient than the segmented market populated by a small subset of agents. Hence, greater genetic diversity leads to greater consistency with fundamental values and market efficiency.

Test Case Evaluation and Input Domain Reduction strategies for the Evolutionary Testing of Object-Oriented software

In Evolutionary Testing, meta-heuristic search techniques are used for generating test data. The focus of our research is on employing evolutionary algorithms for the structural unit-testing of Object-Oriented programs. Relevant contributions include the introduction of novel methodologies for automation, search guidance and Input Domain Reduction; the strategies proposed were empirically evaluated with encouraging results. Test cases are evolved using the Strongly-Typed Genetic Programming technique. Test data quality evaluation includes instrumenting the test object, executing it with the generated test cases, and tracing the structures traversed in order to derive coverage metrics. The methodology for efficiently guiding the search process towards achieving full structural coverage involves favouring test cases that exercise problematic structures. Purity Analysis is employed as a systematic strategy for reducing the search space.

Epigenetic programming: Genetic programming incorporating epigenetic learning through modification of histones

We present the results of our work in simulating the recently discovered findings in molecular biology regarding the significant role which histones play in regulating the gene expression in eukaryotes. Extending the notion of inheritable genotype in evolutionary computation from the commonly considered model of DNA to chromatin (DNA and histones), we present epigenetic programming as an approach, incorporating an explicitly controlled gene expression through modification of histones in strongly-typed genetic programming (STGP). We propose a double cell representation of the simulated individuals, comprising somatic cell and germ cell, both represented by their respective chromatin structures. Following biologically plausible concepts, we regard the plastic phenotype of the somatic cell, achieved via controlled gene expression owing to modifications to histones (epigenetic learning, EL) as relevant for fitness evaluation, while the genotype of the germ cell corresponds to the phylogenesis of the individuals. The beneficial effect of EL on the performance characteristics of STGP is verified on evolution of social behavior of a team of predator agents in the predator–prey pursuit problem. Empirically obtained performance evaluation results indicate that EL contributes to about 2-fold improvement of computational effort of STGP. We trace the cause for that to the cumulative effect of polyphenism and epigenetic stability, both contributed by EL. The former allows for phenotypic diversity of genotypically similar individuals, while the latter robustly preserves the individuals from the destructive effects of crossover by silencing certain genotypic combinations and explicitly activating them only when they are most likely to be expressed in corresponding beneficial phenotypic traits.

DOM/XML-based portable genetic representation of the morphology, behavior and communication abilities of evolvable agents

This article presents the results of our work on the role of genetic representation in facilitating the quick design of efficiently running offline learning via genetic programming (GP). An approach using the widely adopted document object model/extensible mark-up language (DOM/XML) standard for the representation of genetic programs, and off-the-shelf DOM-parsers with built-in application programming interface (API) for manipulating them is proposed. This approach means a significant reduction in time in the usually slow software engineering of GP, and offers a generic way to facilitate the reduction of computational effort by limiting the search space of genetic programming by handling only semantically correct genetic programs. The concept is accomplished through strongly typed genetic programming (STGP), in which the use of W3C-recommended standard XML schema is proposed as a generic way to represent and impose the grammar rules in STGP. The ideas laid in the foundation of the proposed approach are verified by the implementation of GP in the evolving social behavior of agents in predator–prey pursuit problems.

Strongly Typed Genetic Programming

Genetic programming is a powerful method for automatically generating computer programs via the process of natural selection (Koza, 1992). However, in its standard form, there is no way to restrict the programs it generates to those where the functions operate on appropriate data types. In the case when the programs manipulate multiple data types and contain functions designed to operate on particular data types, this can lead to unnecessarily large search times and/or unnecessarily poor generalization performance. Strongly typed genetic programming (STGP) is an enhanced version of genetic programming that enforces data-type constraints and whose use of generic functions and generic data types makes it more powerful than other approaches to type-constraint enforcement. After describing its operation, we illustrate its use on problems in two domains, matrix/vector manipulation and list manipulation, which require its generality. The examples are (1) the multidimensional least-squares regression problem, (2) the multidimensional Kalman filter, (3) the list manipulation function NTH, and (4) the list manipulation function MAPCAR.