Most investment decisions are no longer made directly by humans. Instead, computer programs buy and sell shares of companies. Algorithms, which are essentially a set of predefined rules that determine whether it is the right time to purchase a stock or book a profit, complete transactions at the speed of light.
Over the years, algorithms have come to dominate stock markets. Currently, machines, not humans, carry out an estimated 80% of stock transactions in the U.S.
Why have the big financial institutions switched over to algorithms? The automated buying and selling of shares provide several benefits:
- Speed: Powerful computers are programmed to analyze different markets and execute transactions in a fraction of a second. High-frequency trading is a profitable business as it gives the trader an advantage over others who are slower to react to new developments.
- Stock purchases and sales can be made at the targeted prices, and share transactions are executed at the desired level. There’s a lower probability of losing the chance to make a profit as the trade is executed almost instantaneously.
- Elimination of human error: As a computer executes the transaction, there’s no possibility of a “fat finger” error.
What’s a “fat finger” error? Here’s an example — In 2018, an employee of Samsung Securities, a brokerage that is part of South Korea’s Samsung group, mistakenly issued 2.8 billion shares valued at $104.8 billion to the company’s employees.
The individual who made the error was trying to pay a dividend to the staff. Instead, the computer entries resulted in the issue of new shares. Some staff members took advantage of the error and sold the stock immediately.
While algorithms don’t make human errors, they do slip up occasionally. And when things go wrong, the repercussions can be enormous.
The flash crash of 2010
On May 6, 2010, stock prices in the United States witnessed a remarkable level of volatility. The prices of almost 8,000 shares and exchange-traded funds (ETFs) fell by 5%, 10%, or 15%, and then rose by a similar percentage within the same day.
The price gyrations were unprecedented. More than 20,000 trades occurred at valuations that differed by 60% from the prices that these securities were traded at moments before. In some instances, shares exchanged hands at a penny, while in others, the trades took place at $100,000.
However, things settled down by the end of the day. The market closed at a level that was about 3% less than the previous day.
What happened? Why did share prices move from one extreme to the other in such a short period? How did the market regain its losses?
A joint investigation by the U.S. Commodity Futures Trading Commission (CFTC) and the U.S. Securities & Exchange Commission (SEC) blamed algorithms. The report, which was published five months after the May 6, 2010 “flash crash,” revealed that one transaction by a trader on Chicago’s derivatives market led to the wild market swings.
The report didn’t identify the company that caused the crash, but officials indicated that it was Waddell & Reed Financial, an asset management and financial planning company based in Kansas.
What exactly did Waddell & Reed do to wipe out $1 trillion of market capitalization temporarily?
It seems the asset management company initiated a sell program for 75,000 E-Mini contracts valued at about $4.1 billion. An E-Mini contract is a futures contract or an agreement to buy or sell a financial security at a fixed price on a specific date in the future. The unique aspect of this $4.1 billion transaction was that an automated algorithm executed it. There was no human intervention involved at all.
At this stage, the reader needs to know that this was not the first time that Waddell & Reed was carrying out a transaction of this size. Earlier it had executed a similar sell order for 75,000 E-Mini contracts. But at that time, the order had been carried out by a combination of manual trading and automated algorithms. This transaction had taken over five hours to complete.
But the May 6 transaction, which was completely automated, took only 20 minutes. This sale, which took place over a relatively short period, triggered the algorithms of other high-frequency traders and led to additional selling pressure.
Algorithms played a crucial role in bringing about the flash crash of 2010. The sale of 75,000 E-Mini contracts led to a “hot potato” effect. At one stage of the crash that lasted only 14 seconds, contracts exchanged hands between high-frequency traders 27,000 times. The actual number of contracts that changed hands at the end of these 14 seconds? Only 200.
The downward trend corrected itself when the Chicago Mercantile Exchange (CME) suspended trading for five seconds. When trading resumed, prices began to recover to their earlier levels.
Algorithmic investing – how it works
The flash crash of 2010 demonstrated the dangers of using algorithms. But computer programs that follow a defined set of instructions can serve a beneficial purpose, too.
Long-term institutional investors use them extensively. These financial institutions, which include mutual funds, pension funds, and insurance firms, have a time horizon that stretches for years. Why would they want to use an automated system to buy or sell shares?
There are occasions when, say, a pension fund may want to make a substantial investment in a specific stock. Placing an order for the entire quantity could drive the price up. The other option could be to track the stock’s price manually and place an order whenever the market value is below the threshold that has been decided upon.
Buying all the shares in one lot isn’t a good idea. The pension fund could end up paying far more for the shares than is necessary. Placing multiple manual orders is an inelegant solution. There is the possibility of human error, as well as the chance that the financial institution won’t be able to get the best price for each batch of shares.
The best solution for the pension fund is to buy the shares using an algorithm. This automated mechanism ensures error-free trades and the best prices.
Investors looking for short-term gains also use algorithms. Many institutions adopt trend-following strategies. These work by buying and selling shares based on price trends. For example, a trader could use an algorithm to buy a particular stock when its 50-day moving average price exceeds its 200-day moving average price.
A moving average price helps to identify the trend in prices. Thus, if a 50-day moving average is higher than a 200-day moving average, it indicates that there is a more considerable increase in the price of the share over the last 50 days when compared to the upward movement over a more extended period of 200 days.
There are numerous other trading strategies that algorithm traders could use to make short-term gains.
There is also another category of traders for whom algorithms are of crucial importance — the high-frequency trader.
There’s money in milliseconds.
High-frequency traders (HFTs) use algorithms to analyze markets and spot trends before others can. With this information, they can be the first to place a buy or sell order, allowing them to be the beneficiary of a trend that others are not yet aware of.
There are two critical principles behind the success of HFTs. These are their ability to identify trends early and speed in the execution of contracts.
These traders can execute orders in milliseconds or even microseconds. A thousand milliseconds make a second, while a thousand microseconds make a millisecond. Some HFTs place their computers within the same premises as the exchange. The proximity of the HFT’s servers with those of the exchange allows signals to move between them faster. Closer proximity can give the trader a decisive advantage over the competition.
However, some people think that this is unfair. Why should some traders have faster access to the exchange’s servers? In fact, in a recent judgment, a Manhattan federal court has ordered seven U.S. stock exchanges to justify their action of providing special facilities to high-frequency traders. These included the ability to obtain an increased level of data and permission to locate their servers close to the exchange’s computers.
The class-action suit alleges that ordinary investors were at a disadvantage because they didn’t have access to these facilities. Additionally, they were never told that HFTs were provided with these services. The defendants, who are accused of “fraudulent conduct,” include Nasdaq and the New York Stock Exchange.
Putting the brakes on algorithm traders
Over the years, revenues from high-frequency trading have been declining. According to data compiled by consultancy firm TABB Group, they were at their highest in 2009 when they reached a level of $7.2 billion. By 2017, revenues from high-frequency trading had fallen to $1 billion.
High-Frequency Trading: revenue from US equities ($ billion)
The reasons for this fall in revenue include the fact that the field is getting crowded. A higher number of HFTs means that there is less business to go around. Lower volatility and higher costs have also contributed to the fall. Co-location fees have been hiked. According to Deutsche Bank, these fees “doubled or tripled” in the years between 2010 and 2015.
There’s more bad news for HFTs. Several exchanges in the U.S. and Europe are planning to install “speed bumps” so that the playing field is more level. This would allow traders who don’t have access to the powerful computers used by HFTs, or those without servers at co-location facilities, a better chance of making profitable trades.
Banks and pension funds would be among the financial institutions that would benefit when speed bumps are installed. In recent years, they have been complaining that HFTs have been eating into their profits.
The best of both worlds – machine learning meets human intelligence
Algorithms have allowed high-frequency traders and other investors to profit by automating their transactions. However, some investment professionals are using algorithms in an entirely new way.
Havelock London, an investment management company with an AUM of $23 billion, promises to mix “traditional investment management with modern data science and technology.” The Financial Times describes this as an effort to code a robotic Warren Buffett.
The so-called “Oracle of Omaha” is possibly the world’s greatest long-term investor. Shares in Warren Buffett’s Berkshire Hathaway, a multinational conglomerate, have grown faster than the S&P 500, which is an index of 500 major U.S. companies, by about 2.5 million percentage points over the last 54 years.
What if it were possible to create an algorithm to replicate this performance?
That’s precisely what Havelock is trying to do. The asset management company’s investments are focused on a small group of 38 firms. New companies are added at the rate of one per month. Chief executive Matthew Beddall says he is trying to create a system that resembles a “computer-powered private equity firm.”
How does Havelock use algorithms to develop a long-term investment strategy? It carries out a detailed review of the target company’s performance and arrives at a valuation using both human knowledge and algorithmic analysis. After creating the valuation model, investment activity is almost entirely automated.
How has Havelock’s “LF Havelock Global Select” fund performed? It has provided a return of 5% since its launch in August 2018.
LF Havelock Global Select – Performance since launch August 2018 to September 2019
The bottom line
Algorithms play a crucial role in the investment world. HFTs, banks, pension funds, and several other types of financial institutions use them to boost their profits, reduce costs, and eliminate the possibility of human error when carrying out transactions.
Now, some asset management firms are taking the next logical step. They are trying to create algorithms that mimic human intelligence and will have the capability of identifying long-term investment opportunities. Will these investment algorithms succeed? Only time will tell.
However, one thing is sure. Algorithms shouldn’t be created and then left to function without human supervision. If traders and investors allow them to run independently, the results can be catastrophic. The 2010 flash crash provides ample evidence of this.