Algorithmic Trading Strategies, Explained

Algorithmic trading looks to remove the human factor and instead follows predetermined, statistics-based strategies that can be run 24/7 by computers with minimal oversight.

Computers can offer multiple advantages over human traders. For one, they can stay active all day, every day without sleep. They can also analyze data precisely and respond to changes in milliseconds. To top it off, they never factor emotion into their decisions. Because of this, many investors have long since realized that machines can make excellent traders, given that they are using the correct strategies. 

This is how the field of algorithmic trading has evolved. While it began with computers trading in traditional markets, the rise of digital assets and 24/7 exchanges has brought this practice to a new level. It almost seems as if automated trading and cryptocurrencies were made for each other. It’s true that users will still have to work out their own strategies, but when applied correctly, these techniques can help traders take their hands off of the wheel and let mathematics do the work.

The main philosophies behind most algorithmic trading revolve around using software to spot profitable opportunities and jump on them faster than a human could. The most common practices are momentum trading, mean reversion, arbitrage and a variety of machine-learning strategies.

Most algorithmic trading strategies center around identifying opportunities in the market based on statistics. Momentum trading seeks to follow current trends; mean reversion looks for statistical divergences in the market; arbitrage searches for differences in spot prices across different exchanges; and machine learning strategies try to automate more complex philosophies or integrate several at once. Not one of these is a simple guarantee for profits, and traders will have to understand when and where to implement the correct algorithm, or “bot.” 

Generally, bots are tested against historical market data, which is called backtesting. This allows users to try out their strategy in the actual market they plan to unleash it on, but with established movements from the past. Some risks in doing this can include “overfitting” — which is when a bot is devised around historical data that doesn’t really reflect current conditions, thus leading to a strategy that fails to actually produce. A very simple example would be if you designed and tested a bot against data from a bull market but began running it live in a bear market. Obviously, you won’t see the returns you were expecting.

Momentum trading is based around the logic that if a predominant trend is already visible in the market, then that trend is plausibly going to continue at least until signals begin to come in that it has ended.

The idea with momentum trading is that if a certain asset has been moving primarily in one direction for, say, several months, then we can safely assume this trend will continue, at least until data starts to show otherwise. Therefore, the plan will be to buy on every dip and lock in profits on every pump, or vice versa if shorting. Of course, traders need to be aware of when a market shows signs of trend reversals, or else this same strategy could begin to turn around pretty fast.

It should also be noted that traders shouldn’t set strategies that try to buy and sell on the actual lows and highs, or what is called “catching the knife,” but rather lock in profits and buy back in at levels that are reasonably safe. Algorithmic trading is ideal for this, as users can simply set percentages they feel comfortable with and let the code do the rest. This technique on its own, however, can be ineffective if a market is moving sideways or so volatile that a clear trend has not emerged.

One excellent indicator for watching trends is moving averages. Just as they sound, a moving average is a line on a price chart that shows the average price for an asset over x amount of days (or hours, weeks, months, etc.). Often, amounts like 50, 100 or 200 are used, but different strategies look at different time periods in order to make their trade predictions.

Generally, a trend is thought of as strong when it stays well above or below a moving average — and weak when it approaches or crosses over the MA line. In addition, MAs based upon longer time periods are generally given a lot more weight than one that only watches, say, the last 100 hours or a similar timeframe.

Mean reversion refers to the fact that, statistically, the price of an asset should tend back toward the historical average price. Extreme deviations from this price imply overbought or oversold conditions and the likelihood of a reversal.

Even for something like Bitcoin (BTC), which has really only ever been in a bear market, there can be notable highs or lows that stray from the trajectory the price has historically followed. More often than not, markets will trend back toward this mean price before long. By watching the long-term averages, algorithms can safely bet that massive deviations from these prices are likely not to last for long and set trade orders accordingly. 

For example, one specific form of this is called standard deviation reversion, and it is measured by an indicator called Bollinger Bands. Basically, these bands act as upward and downward limits on deviations from a central moving average. When the price action moves toward one of these extremes, odds are high that a reversal toward the center is coming soon. 

Of course, one of the biggest risks here is that the algorithm can’t account for changes in fundamentals. If a market is crashing due to some flaw in the underlying asset, then it is possible the price will actually never recover — or at least not swiftly. This is, again, where traders need to monitor and account for certain conditions that their algorithms cannot see.

Another form of mean reversion can occur across multiple assets, and utilizing this technique is called pairs trading. Let’s say, two assets are traditionally correlated. That is, when one goes up or down, then statistically, so does the other. An algorithm can be crafted to watch for one of these assets to make a move, then place a trade based on the likelihood that the other commodity will soon follow. The timeframes for these discrepancies can sometimes be rather short, making the automated nature of this strategy far more valuable.

Arbitrage is a strategy that takes advantage of a price difference on the same asset across multiple markets.

Sometimes the same product, like a commodity or currency, can temporarily have different prices on different exchanges. This can offer a great opportunity to make a profit for those fast enough to trade between these markets before they balance out. To this end, an algorithm can be developed to watch various assets across different markets and open trades as soon as discrepancies are found. 

This technique isn’t overly complex, but the traders who can respond the quickest have a distinct edge over the ones who are slower. This is one strategy where high-frequency trading definitely has a notable advantage, as it is precisely the traders taking advantage of these market conditions that will cause the gap in prices to collapse.

Machine learning and artificial intelligence stand to push algorithmic trading to new levels. Not only can more advanced strategies be employed and adapted in real time but new techniques like Natural Language Processing of news articles can offer even more avenues for getting special insight into market movements.

Algorithms can already make complex decisions and make them according to predetermined strategies and data, but with machine learning, these strategies can update themselves based on what is actually working. Instead of just “if/then” logic, an ML algorithm can assess multiple strategies and refine the next trades based upon the highest returns. While they still take work to set up, this means traders can have faith in their bot even as market conditions evolve beyond initial parameters.

One popular type of ML strategy is called naive Bayes. In this technique, learning algorithms make trades based on previous statistics and probability. For example, historical market data shows that Bitcoin goes up 70% after having three consecutive days in the red. A naive Bayes algorithm would see that the last three days have all been down and automatically place an order based on the likelihood it will rise today. These systems are highly customizable, and it will be up to every trader to set their own parameters for things like risk and reward ratios, but once you are happy with a balance, you can let it run with minimal interference.

Another benefit of ML is the ability for machines to be able to read and interpret news reports. By scanning for keywords and having the appropriate strategies lined up, these types of bots can make trades within seconds when positive or negative news breaks. Obviously, these will only be as accurate as the logic that goes into them — and are thus tricky to implement — but still offer an edge over other traders when properly set up.

Note that this is the cutting edge of a new branch in automated trading. So, bots designed to work this way may be harder to find, cost more to access or simply be less predictable than some of the more time-tested techniques.

Order chasing is the practice of watching for certain, very large, orders and then trying to move swiftly based on the assumption that this will lead to further price movement.

Usually, being able to anticipate a large order from a major player would require inside information of some kind, and trading with such knowledge is generally illegal. However, some high-frequency traders have found legal ways to scrape data from over-the-counter trading forums called “Dark Pools.” These types of trading forums don’t have to submit their order data in real time like an exchange, and so their movements tend to have a delayed effect on the market. By gathering and implementing this data faster than the average trader, users of this technique can have a serious advantage over those who don’t.

For example, you see a massive sell order being executed on a Dark Pool. This tells you that soon when this data gets posted to the rest of the market, a great many smaller sellers will probably respond with their own orders. Since this can be anticipated, you can get ahead of the wave and be among the first to sell, which means you can easily buy back in when the dip cools down. Again, this method is not illegal so long as the data is collected through the correct channels, and many algorithmic traders have made this their strategy of choice.

There are many websites that offer a variety of trading algorithms, which you can then connect to the digital asset exchange of your choice.

Quite a few services exist that can get you quickly set up with algorithmic trading. Sites such as TradeSanta, Bitsgap and Cryptohopper all offer multiple types of accounts that can range from free to rather pricey, depending on what tools are made available. For beginners, a free account will generally offer plenty of options to get started, but paid accounts can be very useful if you look to become a professional.

These sites will generally offer tutorials and other material so you can become educated on finding the bots and strategies that are right for you. While not every service is compatible with every exchange, you will find that the majority of these products will support almost all of the largest and most popular exchanges. Some even have special promotions for using their bots in connection with a specific platform, so users should have plenty of options to choose from.

There are admittedly many more techniques and services you can explore, but this guide should give you the basics you need to go out there and get your feet wet with algorithmic trading. Go slow and learn everything you can, and it shouldn’t be long before you decide if an automated strategy is right for you.