How Far Can Algorithms Take Forex Traders?

Forex trading evolved alongside technology. Today, many top traders combine programming skills and reasoning to create algorithmic trading, machine learning, and artificial intelligence. How far can technology take forex traders?

This article was written in consultation with Eduard Samokhvalov, Founder & Developer of Algorithmic Systems at AlgoMaschine.

It wasn’t all that long ago that commentators sat on television explaining to viewers this new concept called ‘email.’ Most people forget that the beloved iPhone is a bit more than a decade old. They take for granted our modern wonders that, unknowingly, come off with an air of entitlement. Yet, many of us saw this revolution in our lifetimes.

The early ‘90s experienced a transformation in the technological landscape that forever changed the world. The speed of information and computational power supplanted paper and pencil. We were no longer bound by the hours in the day but by the speed of our processors.

Today’s market looks light years ahead in comparison. We’re on the verge of cars that drive themselves, let alone facial recognition. Computer technology inches closer to opening the quantum computing landscape.

And yet global finance lags far behind other sectors…the very industry that facilitates global supply chains. It’s ironic given it finances the very companies that changed the world and enabled worldwide commerce. Only recently have startups begun foraying into the world of banking. Still, few players exist in the foreign exchange world.

Since the significant deployment of the MetaTrader 4 platform in 2005, most platform improvements to Forex trading have been incremental. True, algorithmic trading found a considerable following, and substantial capital is spent every year developing new systems. Still, the progress made falls well short of perceptions and expectations. We won’t rely on artificial intelligence to harvest profits consistently anytime soon on a global scale.

We must recalibrate our expectations, understanding where we came from, our current state, and our true potential. Only then will we unlock the true potential technology offers.

The basics of algorithms

This topic confuses readers across the globe. Most don’t understand their true nature and how they relate. Indulge me as I give you the 10,000-foot view of these topics.

Both start with two essential components: math and programming.

Programming enables us to calculate outcomes much faster than pen and paper. The benefits grow exponentially the more difficult the calculations. Since the dawn of mathematic scholars created shortcuts to circumvent laborious calculations. It is these shortcuts that enable what we know as machine learning and artificial intelligence today.

The first handheld calculator hit the markets in 1970. It wasn’t something Isaac Newton could use, nor even Einstein.

Let’s start with one of the most basic algorithms:

Y = MX + B

Students learn this formula years before they learn to drive. It’s the fundamental equation to find the slope of a line. It reads:

Given an input of X, I can tell you the output Y.

Think of a store that prices apples. The boss says the price of apples will be triple the price of oranges and then add on an additional $5.00. So, if oranges cost $2.00, the apples cost $11.00. The equation works as follows:

Y = $2.00 x 3 + $5.00 = $11.00

Here’s the exciting part…every equation that works from algorithmic trading to machine learning to artificial intelligence uses similar equations, just more complicated. An algorithm is a fancy word for an equation, nothing more.

Basic algorithmic trading uses static equations to trade. It assumes that a known equation doesn’t change. The equation will take a variety of inputs to make decision outputs. For example – you buy when price crosses the lower Bollinger Band and sell when it hits the upper Bollinger Band. This strategy is fixed and easily programmable.

Evolution to machine learning and artificial intelligence

Machine learning assumes you don’t know the best equation. Instead, you allow the computer to figure it out for you. Imagine an equation like the one above, but with enough components to cover an entire page. The computer will tweak those values to come up with the best possible algorithm.

Here’s where the mathematical shortcuts come in. The computer won’t know the best equation, but it wants to find out. So, you give it a bunch of inputs and profit outputs as training data (this is known as supervised learning). The computer looks at the data and tries to come up with an equation (algorithm) that uses the inputs to correctly predict the outcome (or get as close as possible).

Immediately, you realize this requires far too many possible scenarios to test. So, you use a mathematical shortcut. Computers try a bunch of points at random. Then it works nearby points. It uses calculus to measure the rate at which they’re improving towards the answer. After that it’s rinse and repeat.

Here’s an easy way to think about it. Picture a trampoline with multiple bowling balls on it at different points. Each of them creates an indent in the flat fabric. The lowest points created by the bowling balls in the fabric represent solutions.

Artificial intelligence is a branch of machine learning which attempts to mimic human intelligence. It focuses on the decisions rather than the outcome. Machine learning works with a subset of targeted choices. Alternatively, artificial intelligence considers a more holistic perspective.

Think of it this way. Machine algorithms trade specific market conditions such as oversold or trending. Artificial intelligence would try to manage both the trading and overall portfolio while predicting the future. With driving a car, machine learning solves driving through traffic lights. But artificial intelligence ties together all the different decision-making scenarios we make to drive you to your destination.

Or simply put – artificial intelligence brings together multiple machine learning algorithms to replicate how we adjust to our environment.

Slow acceptance within Forex

Machine learning turned up marginal results at best. While algorithms solve specific parameters, they fail to adapt to changing market conditions. Jumping from one pair to another might as well be a foreign language. This limits much of our current capacities to high-level systems that deliver slightly better than market average results.

Yet artificial intelligence remains elusive. Your slowest operation dictates the speed of your value chain. No decision works faster than the most time-consuming preceding bottleneck. Elon Musk’s self-driving cars require an inordinate amount of data to solve the simplest tasks. Ironically, the transactional data for Forex is quite small relatively speaking. Creating complex trading systems requires data not readily available.

Traders instead work outwards from their hubris. Successful traders attempt to recreate their abilities. But they either fail to articulate their reasoning or are simply limited by programming capabilities. Unicorns remain elusive, these few people that straddle the line of programmer and trader. It’s only through true collaboration of open-source sharing that we make progress.

Programming relies entirely on data that is…well, programmable. It sounds obvious. But you cannot expect a computer system to synthesize multiple economic papers and news articles to deliver comprehensive ideas. They’re limited by inputs and outputs. Systems have no place for thoughtful analysis of central bank statements. Sure, they devour technical indicators with ravenous hunger. But they cannot comprehend political philosophy and consequences.

Practical applications

The simplest algorithms free resources to exploit other opportunity costs. They create predictive analytics that enables better decision making. AI developers found more success developing predictive analytics rather than trading programs. Decoupling the profits from decisions creates healthier systems. That follows advice from successful traders that stress focus on decision making rather than outcomes.

Foolish traders dismiss the technology outright based on these criticisms. It’s utility not only reduces workload but creates synergies that otherwise remain dormant. Programmers often start by creating alerts or reducing the search for trading ‘setups.’ That frees up time to work other opportunities.

Numerous traders continuously create profitable algorithms. They may fail in different market environments. But, that’s not much different than a manual trend trader whose strategy only works when a market trends in one direction. Some rely on multiple systems working in concert to develop holistic systems that perform significantly above market averages.

Traders balance marginal returns with effort. A system that wins 67% of the time with a one to one risk to reward ratio with low maintenance might be optimal. Doubling the effort may only yield marginal returns. Most systems need constant adjustment, with few working across all market environments.

Alternate market conditions create the most challenging obstacle for programmers. How do you explain to a computer the intricacies of geopolitical tensions or introducing negative interest rates? Most of us fail to understand these concepts on our first pass. Assuming a computer intelligence could do better is foolish. Rather, use systems to create the foundations for basic strategies that you overlay your nuances.

However, the real benefit lies in reducing workload or creating marginal statistical systems. Simply put, computer applications aren’t capable of the same capacity as human intellect. Instead, we’re limited by the computational power, technology available, and its capability to replicate the intricacies of our decision making.

Instead, working on alert systems that focus our attention creates the best benefit for most traders. This requires the least amount of programming knowledge, yet still delivers significant impact. Think of a trader who relies only on a handful of setups to produce profits. He identifies the specific parameters that define his entry or passes on a trade. Now, picture that trader only turning towards the charts as the alerts require. That’s how algorithms deliver power.

The irony of the solve

Solving the perfect AI only creates a new playing field that shares the same problems. Information disseminates, leaving AI systems competing against one another. That brings us back to our current environment.

Information arbitrage lost much of its teeth in the last few decades. The speed of information quickly disseminates information across the globe. The smartphone quickly became replicated within a few years. Most technological advances only acquire limited patent protection. Afterward, they bare themselves to the world’s pillaging.

Any perfect algorithm or intelligence that pilfers markets of profits will quickly be eclipsed. By its very nature, efficient markets cannot tolerate such loss. On a large scale, it would destabilize global exchanges. Instead, any large scale progress would quickly disseminate among the traders.

That creates a feedback problem. If multiple systems solve for the same outcome, what creates the profit stream? Is it the faster solve or the one that diverges the most? At that point, we then create AI alternatives to mimic human behavior, creating an alternative, yet zero-sum scenario of our current state.

Drawbacks to Forex

Scalpers and traders get more immediate benefits from current AI trading systems. Some are traders that seek and exploit pricing arbitrage within the system to the benefit of themselves. These areas typically die out quickly. Yet, it’s questionable whether they add value to the market.

AI works wonderfully in spotting patterns among noise. That works exceptionally well when trading breaks down to market depth and order books. Information is abundant at this level. It allows their systems to quickly analyze and adapt to the conditions that rely less on qualitative analysis.

A return to what matters

In the end, technology reminds us of our humanity. The digital age pulls us away from the very traits that create profitable traders. We need a convergence of our ideals and broker capabilities. Any broker can connect you to simple trading and execution without knowledge of you or your goals.

Ironically, that very arms-length transaction limits profitability among traders and investors. Consider the numerous persons and companies that went bankrupt through inept management. It’s the trust and integrity of our partners that check our resolutions.

That’s where long-term relationships work to our benefit. Companies that focus on their customers and their needs holistically build back the trust once lost. It’s a combination of delivering transparency, quality execution, and collaborative efforts. 

A programmer that delivers a billion-dollar trading system without a trusted broker is worthless. Its our partners that hold the keys to our success as traders and investors. So consider what’s important – algorithms will only take you so far now and forever.

Photo: @franckinjapan

Recommended Posts