Understanding the validity of relationships in machine learning model inputs and the predicted variables is essential in the realm of algorithmic trading. To achieve meaningful predictions, the relationships within the data must be real and not just statistically coincidental. This discussion highlights the importance of employing unit root tests on model residuals to check for spurious regressions - models that falsely indicate meaningful relationships due to coincidental statistical errors rather than genuine interdependencies.

Such models can misleadingly show low error metrics, giving a false sense of accuracy and potentially leading to costly mistakes in trading strategies. The text discusses generating synthetic time series data to illustrate how spurious regressions can occur and provides insights into identifying and mitigating them using statistical tests like the Augmented ...

Read more...

A new script has been developed that facilitates the conversion of historical data across multiple trading timeframes including M1, M5, M15, M30, H1, H4, D1, W1, and MN. The process begins by downloading or importing your historical data.

Once data is imported, it is essential to close all charts and restart the MT4 platform. From there, users must navigate through the menu to File -> Open Offline and open an offline chart of the target symbol in M1, which should be marked in grey.

The script, named "period_converter_all_kai," can then be executed from the script folder. Users will need to input a "Start Date" and press OK, then wait for a confirmation message indicating the successful conversion. Restarting the MT4 platform will allow users to access and verify the data across all desired timeframes.

For users seeking a more detailed walkthrough of this process, complete with visu...

Read more...

In our ongoing development of a manual EA, previously discussed, we have introduced considerations for moving towards a universal EA model. The creation of the C_Manager class marks a strategic update—it functions as an intermediary layer between the EA and the order system, enhancing security and controlling functions. This class acts as an administrative point, increasing reliability by managing essential operations independently, based on predefined parameters.

The transition introduces a structured approach where the EA accesses order functions through C_Manager, which performs operations under strict conditions to reduce error risks. This means actual trading decisions and operations, like opening market positions, become more streamlined and secure.

Additionally, different account types—NETTING and HEDGING—play critical roles in automatic EA functionality. Each account type ha...

Read more...

Understanding the DD_Relative function in trading software is crucial for monitoring the financial health of a trading account. This function calculates the relative drawdown of the account by determining the current net profit or loss of all open positions. This is achieved through the Current_LossOrProfit() function, which aggregates these values. The net figure is then divided by the account's current balance, sourced from AccountInfoDouble(ACCOUNT_BALANCE), to determine the drawdown as a percentage.

The percentage result is refined to two decimal places using NormalizeDouble(), and displayed on the trading chart via the Comment() function. Additionally, the Current_LossOrProfit() function plays a key role by sifting through each open position, checking its alignment with specific criteria such as matching magic numbers and symbols relative to the chart in focus. It then incorpora...

Read more...

Neural networks, while complex, offer a broad range of applications once mastered. The shift in focus from feature design to architecture, loss function, and optimization design represents a move towards higher abstraction in machine learning tasks. This transformation underscores the need for depth in understanding the intricacies involved in neural network implementation, particularly when applied to fields such as algorithmic trading.

Understanding the components and functioning of neural networks is crucial. Neural networks, inspired by biological neural networks, operate through a structured network of layers comprising interconnected nodes. This structure allows them to identify complex patterns in data, often requiring thorough analysis and careful management of the scaling and preprocessing of input data to ensure model accuracy.

Especially in trading applications, incorpora...

Read more...

A professional Grid advisor that utilizes the Relative Strength Index (RSI) is now available for integration in trading systems. This advanced tool reduces drawdown on accounts by managing unprofitable orders through strategic overlapping. Users gain clear insights into profits via the chart display, enhancing utility for manual trading operations.

The advisor is optimized for use with ECN brokers that offer low spreads, improving execution and potential profitability. It is strongly recommended to initially test this tool on a demo account to familiarize with its functionalities. Additionally, pairing this advisor with Buy Sell Signals can facilitate trend following and allow for semi-automatic operation, where users can specify conditions such as "Buy Only" or "Sell Only".

Key customization options include adjustable RSI periods, bounds, and timeframes. Management features extend ...

Read more...

Understanding Trailing Stops: A Guide for Traders

Trailing stops are critical tools for capitalizing on gains while minimizing risks. They dynamically adjust to price movements in favorable directions but remain static when the trend reverses. Trailing stop mechanisms vary, with platforms like MetaTrader offering integrated options which can also be incorporated into Expert Advisors.

Traders can customize trailing stops using market data, technical indicators, and varying methodologies that intertwine trading strategies with risk management practices. The setup of a trailing stop involves defining parameters that balance potential profit against the risk, accounting for factors like market volatility and transaction costs.

Effective use of trailing stops requires a strategic approach to setting stop loss levels. They should not only cover all incurred costs like swaps and commissio...

Read more...

In the latest analysis of the Modified Grid-Hedge EA, attention is focused on the mathematical dimension of optimizing Grid EA strategies. With an existing framework as the baseline, this detailed examination uncovers the core mathematical strategies that guide the EA's behavior, setting the stage for future code-based enhancements.

This segment does not cover the coding aspects; instead, it delves into the theoretical constructs, formulas, and calculations needed to optimize trading strategies effectively. From a comprehensive recap of Grid Strategy to evaluating parameters like Initial Position, Initial Lot Size, Distance, Lot Size Multiplier, and Number of Orders, the groundwork is laid for quantitatively assessing various trading strategies and outcomes.

Analyzing the profit function for a scenario with sequentially increasing lot sizes reveals intricate relationships between or...

Read more...

The Closed-Form Policy Improvement (CFPI) algorithm offers a new perspective for addressing the instability noted in offline reinforcement learning models which traditionally rely on stochastic gradient descent (SGD). The CFPI algorithm mitigates the need for hyperparameter tuning, often challenging due to limited interactions with the environment, providing a more stable avenue for policy training. By utilizing a first-order Taylor approximation, CFPI constrains the distributional shifts and models behavior policies closely resembling the training data distribution.

Highlighting a key feature, the CFPI method avoids the typical fluctuations in policy performance associated with SDS-based approaches by engaging a deterministic policy shift towards value improvement. The researchers outlined the benefits of modeling behavioral strategies using a mixture of Gaussian distributions. This...

Read more...

Causal inference is fundamental in diverse fields such as econometrics, medicine, and marketing. However, the approach holds particular intrigue for those designing algorithmic trading systems in machine learning landscapes. While ancient philosophers like Aristotle attempted to embed causality into the crux of scientific understanding, modern developments in machinery and algorithms have observed a shift in comprehension. Today, causal inference zeroes in on separating correlation from causation to forge reliable decision-making frameworks.

Machine learning techniques, particularly in the realm of trading, must grapple with causal relationships to enhance prediction accuracy and reliability. The utilization of advanced causal inference methods indicates that although machine learning can interpret data to predict outcomes, discerning the true causal effects remains complicated. This...

Read more...

For traders utilizing MT5, a new script is available that automates order placement ahead of high-impact news events. To deploy, simply double-click the script approximately two minutes before news releases such as FOMC, NFP, bank rates, or CPI announcements. The script is designed to set both Buy Stop and Sell Stop orders swiftly to capitalize on rapid pip movements.

Key features include assignable hotkeys, with the recommendation to use CTRL + 1 for quick access. The script incorporates Stop Loss functions to secure trades if the market moves against the predicted direction, along with a preset Target for potential profits.

Users should note that this tool is advised for use with significant, market-moving ("RED") news only. If the set orders—Buy Stop and Sell Stop—are not activated within three minutes post-announcement, it is recommended to cancel the orders and prepare for the ...

Read more...

The Bacterial Foraging Optimization (BFO) algorithm, inspired by the foraging behavior of Escherichia coli bacteria, serves as a robust technique for addressing complex numerical optimization problems. BFO mimics the natural tactics of E. coli, utilizing mechanisms like chemotaxis—where movement is guided by chemical cues. This algorithm is especially noteworthy in distributed optimization and control tasks across various real-world applications. Utilizing a combination of bacterial behaviors such as swimming and tumbling, the BFO algorithm enhances the search for nutrient gradients, effectively navigating through solution spaces.

Key components of BFO include initialization of a bacterial colony, execution of chemotactic steps where bacteria either swim towards favorable conditions or tumble in adverse environments, and replication phases ensuring survival of the fittest, thus foste...

Read more...

An optimized version of the Hurst exponent indicator has been recently updated for enhanced performance. It is built upon the initial version which has been thoroughly described, providing users with improved computational efficiency. While further optimizations are possible, such as implementing a faster approximation of logarithms, these would result in a trade-off between speed and accuracy. The current improvements focused on significant computational enhancements without drastic sacrifices in precision. This refined tool is crucial for those who rely on detailed quantitative analysis in their programming and technical work.

Read more...

In the realm of optimization, particularly focusing on population-based algorithms, multi-population strategies offer a nuanced approach by employing multiple independent groups to enhance problem-solving capabilities. This method facilitates diverse explorations of the solution space through parallel processing by unrelated populations, while strategically sharing inter-group findings to refine solutions.

Multi-population and multi-swarm techniques, though distinctly powerful, integrate cooperative dynamics within their core framework. Utilizing multiple social groups, or 'swarms', these algorithms foster an interconnected learning environment where individual groups evolve based on shared successes and adapt dynamically to changing problem landscapes. As each group functions with a semi-autonomous strategy, the collective synergy leads to improved performance in finding optimal sol...

Read more...

Goal-Conditioned Behavior Cloning (BC) offers a significant approach for tackling varied problems in offline reinforcement learning, veering from the traditional method of evaluating states and actions. By aligning an agent's actions with predetermined goals under specific environment states, BC leverages supervised learning techniques and historical data to train the agent’s behavior policy. Highlighting the approach, recent papers have shown sequence modeling's role in enhancing policy learning from offline trajectories, posing queries about optimal goal-setting for learning trajectories and devising effective policies.

A key development discussed is the Goal-Conditioned Predictive Coding (GCPC), which integrates sequence modeling into a two-stage framework to refine agent behavior. This innovative model involves pre-training to compress trajectory data into nuanced representations...

Read more...

The latest version of MetaTrader 5 build 4350 introduces new analytical objects to the web platform. Utilize the ruler to measure time and prices, draw shapes (rectangle, ellipse, triangle, and circle), and add labels to your charts.

The new Welcome page in MetaEditor will assist trading app developers. Access educational materials, stay informed with the latest news and monitor your sales.

In addition, Copilot's code completion feature now supports the latest ChatGPT model, GPT-4o.

Read more...

In exploring the development of large language models (LLMs) using CPU-based systems, it's crucial to acknowledge the current landscape where most models rely on Transformers. Software libraries such as 'Transformers' and 'tiktoken' provide robust data processing methodologies that integrate seamlessly with these models. Specifically, tokenizers perform pivotal functions within Natural Language Processing (NLP), converting text into tokens that can be transformed into input vectors understandable by computers.

This post delves into the practical aspects of training LLMs on CPUs, particularly focusing on generating and processing datasets, a critical but often challenging part of model training. It is highlighted that despite the potential limitations of using CPUs—such as the inability to handle complex model functions—various model versions are accommodative of different hardware ca...

Read more...

In the realm of time-sensitive computing operations, validation of user-defined settings is critical. A function effectively manages this by verifying if user-selected time periods are permissible. It accepts two parameters: "allowedPeriods," an array listing the time intervals approved by the system, and "periodsToCheck," which includes the intervals selected by the user.

The processing involves a straightforward validation loop where each entry in "periodsToCheck" is cross-referenced with "allowedPeriods." Should any period from the user’s selection not be listed in the allowed array, the function terminates and returns "false," signaling an invalid or unauthorized choice. Conversely, a consistent match across all entries results in a return value of "true," confirming all user-selected periods are valid.

This method ensures system integrity and adherence to predefined constrain...

Read more...

In this installment of the series on wizard-assembled Expert Advisors, the focus shifts to the incorporation of economic calendar news into the Expert Advisor during testing. This integration aims to either validate a trading concept or enhance the robustness of a trading system. The discussion primarily harnesses the capabilities of the MQL5 IDE tools.

One pivotal aspect covered is the potential trading edge that economic data can bring to a trading system, emphasizing fundamental analysis over technical approaches. Such economic fundamentals include inflation rates, central bank interest rates, and unemployment rates, among others. These elements are crucial as they often trigger volatility in the markets following news releases, with non-farm payroll data being a prominent example.

Moreover, the article explores the utility of SQLite databases within the MetaEditor IDE, proposing...

Read more...

Understanding Data Management and Basic Programming Concepts in MQL5

In programming, particularly in MQL5, understanding functions, algorithms, and data storage is crucial. Functions perform specific tasks based on predefined steps called statements, which include comparisons, repetitions, or data manipulation. Each function operates within an algorithm, a sequence of instructions to solve more extensive tasks. MQL5 supports programming trading actions varying in approach as the same objective can be achieved with different algorithms.

Data in programming is essential and can range from price values to graphical coordinates or sound playback triggers. Data is stored in Random Access in two forms: variables and constants. Variables can change during program execution, whereas constants remain fixed. Knowing these distinctions helps in optimizing program efficiency and debugging.

Spe...

Read more...