Artificial intelligence (AI) trading dominates 60-73% of US trade volume, and it’s not just big bank throwing their weight around. AI is quickly infiltrating small firms and retail investor tools. But AI isn’t a full-size decision making robot. It only brings value in a small set of use cases.
AI trading works by collecting and analyzing vast amounts of financial and alternative data to hypothesize future price movements, automate trade execution, optimize asset allocation, detect fraud, and contextualize market sentiment.
It does NOT have the capacity to automate the entire industry, which still relies heavily on human judgement, perception, and face-to-face client service.
How It’s Used
It has 6 uses, and only 1 is “cool.” It’s the popular predictive trading, where AI analyzes price movements and trades high volumes in an attempt to profit. Lesser known applications include automated trading based on predefined criteria, asset allocation for funds and individuals, and micro-level automations for retail investors like you and me.
#1 Predictive Trading
AI predictive trading refers to the use of machine learning algorithms to buy and sell securities based on price movements. It’s a short-position technique that can be partially or fully automated.
Though not always a high-volume technique, many firms combine it with high-frequency trading (HFT). HFT refers to the purchase and sale of huge volumes of shares in fractions of a second to amplify the returns on small price movements.
HFT also contextualizes market-sentiment. AI uses alternative data sources such as news and social media as inputs in an attempt to better predict price movements.
#2 Automated Trading (Based on Parameters)
AI automated trading is a technique used to remove manual screening for securities that meet predefined purchase or sell criteria A common example is stop losses, which tell a computer to buy or purchase securities when the reach a predefined price.
Advanced forms of automated trading rely on AI to screen securities for factors such as financial ratios and price points and execute trades automatically.
For example, traders may want to buy an oil-sector stock with P/E less than 10 and a EV/Sales of 6 when the TTM price evolution is -30% or more. Instead of manual finding these deals, AI does it for them. Advanced strategies such as deep value investing are the biggest beneficiaries.
#3 Asset Allocation
Asset allocation is the practice of diversifying positions based on risk tolerance and investing goals. Artificial intelligence helps asset managers find the right mix for their clients. This approach is lesser known because it doesn’t aim to maximize profits as fast as possible. It’s purpose is to maximize the chance that clients reach their specific goals.
For example, a 68 year old retiree likely has a lower risk tolerance than a 25 year old entrepreneur. AI asset allocation takes input from clients and creates bespoke portfolio recommendations.
#4 Fraud Detection
AI Fraud Detection works the opposite way as predictive trading. It scans for trading activity that suggests a trader may have access to insider information. For example, if a trader purchases large amounts of a security right before the company announces a massive stock repurchase plan, this indicates insider information.
In addition, advisors working on commission might aggressively buy and sell a position to rack up fees. This is market manipulation, and AI can scan activity at scale to identify illegal tactics like this.
#5 Retail Investor Tools
Here’s where it gets interesting. AI trading tools are also available to retail investors. We have access to number 2 and 3 above (automated trading and asset allocation). You will find them marketed as stock screeners, signal trade routers, and basic digital brokers. Even some alternative data sources are available, like Bright Data.
Retail investors can also invest in AI-enhanced funds.
#6 Combinations (Funds)
There’s another option for everyday investors. It combine prediction, automation, and asset allocation, all done by someone else. Most people don’t think of Exchange Traded Funds (ETFs) this way because they operate much like normal securities from an execution standpoint, but there are a few proud funds running entirely on AI that have seen success over time.
How It Works (3 Ways)
AI trading uses machine learning. Machine learning is a technical way of saying a computer that replicates human intelligence.
What do humans do? We receive sensory input, process it, and come to conclusions. Some subjects have fixed parameters like math while others are subjective and rely on trait categorization, like knowing a dog from a cat.
Machine learning does the same thing. It looks at loads of data and either knows the answer in fixed-parameter scenarios or identifies patterns in subjective scenarios. Two “types” of ML are supervised and unsupervised. The former just means humans tell the computer what output to provide when it recognizes a given input, and the latter lets the computer find patterns of its own.
The tricky part is applying patterns to use cases. Trading use cases fall under trend recognition (technical analysis) or value recognition (fundamental analysis), and the AI works differently for each. The first is unsupervised and the second is supervised.
#1 Unsupervised AI for Technical Analysis
Machine learning traders use a handful of techniques. They look for correlative patterns between real-time macroeconomic data and news against stock prices. They also simply look price trends and attempt to predict price direction based on points of support (lows) and resistance (highs).
The idea is straightforward, but the reality is these techniques produce average results because competing AI counterbalance competitive speed and insight, and because even AI cannot identify order in chaos. Perfect correlations simply do not exist.
#2 Supervised AI for Fundamental Analysis
Supervised AI is a fundamentalist trading tool because it uses predefined criteria to execute a trade. For example, a human trader instructs his AI to scan the market for stocks in manufacturing with P/E less than 10 and P/BV less than 5. This is a simple example and reality is a bit more complex. A trader would feed ML data on P/E and P/BV and ask it to predict price movements.
#3 Bonus: Unsupervised AI for Data Collection
Unsupervised AI also tackes data collection. It scrapes the web for news and social media sentiment data. Many vendors market this service as AI, but in reality it’s web scraping. The AI part comes when ML actually crunches the data (part #1 and #2).
Where It Works (Stock, Forex, Commodities & Crypto)
Whether or not supervised, AI trading doesn’t work for every security. Of the roughly 7 core asset classes, AI works best with stocks, forex, commodities (futures) and crypto.
The reason is these classes have high volatility, undefined return rates, and a reasonably large number of speculators, all of which create a closed system. However, bonds, real estate, and other alternative investments like designer handbags have low volatility and/or not enough quick, speculatory trading to create a reliable market.
(Crypto is the exception here since it’s a decentralized system, but given the volume it’s important to mention.)
How to Do It Yourself in 4 Steps
Step 1: Choose One of Three Tactic
The first step is to decide your technique. This should make sense now that we’ve covered what AI trading is specifically capable of. You only have three options. Either you develop an unsupervised ML model that uses alternative data and price trends, you invest directly in AI-enhanced funds, or you use stock screeners with brokers and trade routers to automate trades based on criteria.
Here’s a breakdown to help you decide:
|#1 Develop Algorithm
|#2 Use Tools to Automate Trades
|#3 Invest in ETFs
Step 2: Set a Time Frame
A defined time frame provides a controlled environment for testing your AI tactic and prevents wishful thinking. A good rule of thumb is to hold positions over at least 3 years, but that doesn’t make sense for algo trading. Here are some potential periods per tactic.
|1 hour (for HFT)
Algorithmic trading attempts to profit from movements in the market, so you can test it in as little as a few minutes, but it’s a good idea to allot at least and hour. It probably doens’t make sense to do more than a year, however, since you’ll want to update and refine the model.
Automated trading should replicate your normal trading activities. If you’re a long investor like me, that means 1 year on average, but short traders are day traders and execute deals in a matter of minutes.
ETFs are long investments, and AI-enhanced funds have beat the market only in very recent and 5+ year periods. We’ve discussed this at length in a number of articles.
Step 3: Program Your Tool
This is where you’ll need to actually create your algorithm, set up your rules for automated trading, or choose your ETFs. Creating an algo is outside the scope of this article, but examples of automated parameters include quality factors such as industry competition and capital allocation to stock repurchase.
Step 4: Backtest
Backtesting consists of acquiring dummy data for past periods and running your configured tool on it to evaluate success rate. I think the reason is obvious. If your model fails on past data, you need to reprogram. It’s a standard model testing practice used in every ML approach.
Step 4: Launch
The scariest part. Launch your algo, activate automated buys/sells, or buy your ETFs. Wait your defined period of time and pull out, then refine to improve performance.
Easy right? Not really. Trading with AI is a tough gig, and it frustrates many traders because like any kind of trading it requires a good amount of trial and error to get right.
Verdict: can AI “predict” the market?
No, AI cannot predict the market. No person or thing can predict the future. However, a handful of AI-enhanced ETFs have outperformed the S&P 500 index over 5-year spans by approximately 5%, and as AI grows in efficacy and popularity, it’ll be more dangerous to dismiss it than learn it.
The world’s capital markets are already full of robots, and the line between “them” and “us” gets thinner by the day.