I’ve been reading the book Deep Thinking by Garry Kasparov. Kasparov is one of the greatest chess players of all time. He ranked number one from 1984 until his retirement in 2005, obtaining a peak rating of 2851.
Kasparov is also remembered as the first world champion to lose a match against a computer, IBM’s Deep Blue in 1997. This result marked the end of human dominance in chess and paved the way for computer domination in other games such as Go and Jeopardy. There are numerous accounts of the Deep Blue match so I won’t go into the details here.
What’s more interesting to me is what happened after the match when Kasparov introduced a new way of playing called ‘advanced chess’ or ‘freestyle chess’. This style of chess gives human players the option to use computer programs to help decide on their next moves.
The idea is that a strong human player might benefit from playing alongside a computer and be able to defeat a super computer playing on its own. In 2005, a tournament was set up that included amateurs, grandmasters and computers. What happened next shocked everyone.
The Winning Team Made Up Of Amateurs
Everybody assumed that the winner of the tournament would either be a computer or a team of grandmasters playing alongside a computer. Instead, an amateur team called ZackS won the tournament, convincingly defeating a team in the final that consisted of grandmaster Vladimir Dobrov, his highly-rated (2600+) teammate plus their computer programs.
The fascinating thing is that ZackS was a team made up of two amateur players with scores of 1685 and 1398 respectively plus 3 computer programs. These two amateurs would never have been expected to win but they found a way to work intelligently with their computer programs so that the sum was significantly greater than the individual parts.
This was a big surprise, even to Kasparov, and prompted him to write a now famous article which included the line:
Weak human + machine + better process” is superior to “Strong human + machine + inferior process.”– Gary Kasparov
The implications of this statement are profound. Amateur players with the help of computers and a strong process can beat even the strongest grandmasters and super computers.
The key insight here is process. ZackS did not have the smartest players or the strongest computers. But they had a methodology that allowed them to harness the best of both worlds.
Building A Process For Chess
So what exactly was their process? According to Nicky Case, the two amateur players ‘would ask the computers what the next move should be and when the computers gave back different answers, they would ask them deeper questions’.
In other words, the amateurs would look for times when the computers disgreed. Then they would work harder with the machines to look deeper into all the possibilities and come up with a more optimal move. This back and forth, allowed the team to more effectively combine the strategic advantages of the human player with the superior processing power of the computer.
According to Tyler Cowen of George Mason University, the results of freestyle chess show that ‘the person working the machine doesn’t have to be an expert at the task’ and that ‘knowing one’s limits is more important than ever’.
In other words, the amateurs were successful because they knew their limitations and therefore, knew when to listen to the computer. A grandmaster, meanwhile, might be more likely to override the computer at the wrong moment.
Implications For Stock Market Investors
Naturally, reading about the success of freestyle chess I could not help but think of the implications for investors. If amateur chess players can beat super computers and grandmasters then ordinary investors can be successful if they have a strong process in place too.
The first key to this, is understanding the strengths and weaknesses of computers versus that of humans. There is no point combining a computer and a human if you don’t harness the advantages of having both.
This brings us neatly to Moravec’s paradox which states that computers are strongest where humans are weak and vice versa. This insight is crucial in order to form a strategic alliance between the two.
Computers, for example, excel at crunching numbers, arithmetic, memory and logic. They are able to process complex calculations at speeds a human can only dream of.
Meanwhile, computers are weak (and humans strong) in areas of creativity, strategy, empathy and intuition. Artificial intelligence is getting better all the time but humans are still better at pattern recognition and putting things into context.
Thinking carefully about Moravec’s paradox I’ve come up with some ways in which investors can work better with intelligent machines. I’ve separated it into two parts; how machines can help investors and how investors can get the most out of their machines.
How Machines Can Help Investors
Let’s look at practical ways that machines can help investors. Machine generally refers to computer and could also refer to wearable technology or things we haven’t yet invented.
#1. By removing errors
One of the main sources of loss for investors comes from making mistakes. As Charlie Munger has said there is immense power from trying to be ‘consistently not stupid instead of trying to be very intelligent.’ And computers can help investors reduce mistakes in a number of ways.
Often investing mistakes are borne out of human biases or emotions getting in the way of good decisions. Biases like overconfidence, misconceptions of chance, anchoring, availability bias and greed. Many of these biases are evolutionary traits that are hard to eliminate.
A computer on the other hand only suffers from bias if it is programmed in. A computer is never going to suffer from overconfidence, become emotional or tired. It will remain objective at all times and present answers that are as truthful as the questions and data that is fed into it. Feed the computer the right data and the right questions and it is going to give you honest, bias-free answers.
It’s important to remember that investing mistakes can also come in the form of omission. In other words, when you miss out on a good investment. Once again, this is where a computer can help by alerting you to all possible opportunities. Also, by keeping track of the investments you missed, investments that were recommended but not taken, you can learn how to not make those mistakes in the future.
Tracking and analysing trades in a computer, particularly losing trades and trades that were missed is a powerful process for eliminating errors and improving trades. Computer programs can also help eliminate errors by alerting a trader to moments where their trading is deteriorating. Health trackers can identify when you are tired and thus more likely to make a mistake and distraction apps can help to keep you focussed on what you’re doing.
#2. By scanning thousands of opportunities
The typical fundamental analyst has an area of focus and is limited to only a small section of the investment universe. This can become a problem when that sector falls out of favour. Similarly, there is no way a human analyst can keep track of all the thousands of stocks that exist on a daily basis. A computer, of course, can scan through thousands, even millions, of different opportunities in just a few seconds.
Most investors are aware of computerised stock screeners which have become an invaluable tool for filtering down the long list of opportunities to a more manageable level. However, fewer investors take the screening approach to another level by truly incorporating it into their process.
Investors might want to ask how their screening is actually affecting their investment results and whether they can come up with more sophisticated screening techniques.
If you screen for low P/E ratios or new 52-week highs, how do you know those indicators are beneficial to your investing approach? Maybe there are more powerful screening formulas you can create which you can apply to a wider investment universe than most investors are looking at.
#3. By backtesting different signals and scenarios
An obvious use for a computer is in the backtesting of trading signals and strategies. Backtesting does not guarantee investing success because the future is never the same as the past. But it’s useful to know what worked in the past and what frameworks have proven to be robust over time.
Through backtesting, an investor can work out market ‘truths’ that are robust over time and develop strategies that provide a successful partnership to human analysis.
For example, in another article I listed the long-term, historical performance of various trading signals. Such trading signals, if robust, provide a profitable base for which to filter down stocks ready for deeper analysis.
My personal approach is to start with profitable opportunities as selected by the computer then use my own skills and experience to delve deeper into these investments and pick the best ones. A computer can also provide useful insights for details like position sizing and when to exit.
Furthermore, you might have an idea for a trading strategy but no idea whether or not it would be profitable or not. By coding and backtesting that strategy you can see whether or not the strategy has any merit and is worth following.
#4. By finding unknown patterns
The human brain is adept at pattern recognition for evolutionary reasons. It’s said that pattern recognition skills were vital for our survival from predators and that a baby is able to recognise a human face at birth.
However, computers are capable of finding profitable patterns on their own, patterns that we might miss.
Using a computer to search through thousands of terabytes of data to find profitable trading signals is often called data mining whereas more sophisticated versions are called artificial intelligence or machine learning.
The reason these techniques work is not because computers are more intelligent than us or better at pattern recognition. It’s simply because their greater processing power allows them to crunch more numbers and bring up more opportunities.
You might want to follow these patterns without further analysis or you may prefer to rationalise a pattern with some kind of logic. Jim Simons is reported as saying he liked data mined patterns because they were less likely to be discovered by other traders.
Computers can also be used to find anomalies that a human investor might miss, perhaps because it is in an obscure market or because it didn’t make the news.
#5. By running real-time calculations
Computers can crunch numbers in real-time faster than any human can. This can provide advantages to day traders who need to make quick calculations on the fly. In fact, some successful traders have revealed how they developed proprietary software to help with their day-to-day trading.
For example, one of the traders in Unknown Market Wizards, uses a piece of software that reads economic releases and automatically calculates whether or not the news is better or worse than expectations. Similarly, Flash Crash trader Nav Sarao used a piece of software to automatically place, execute and cancel bids and offers into the order book.
Both of these traders run the software alongside their trading programs allowing some human interference. They choose when to turn them on, when to turn them off and how much to lean on them. These types of programs form a connection between the machine and the trader and allow synergy in real-time.
Computer programs can also be used to show a live view of risk management and they can be used to alert traders of potential emotional problems such as going on tilt. Restrictions can be built in to stop a trader from trading after a series of losses.
#6. By speeding up the accumulation of experience
In his early years, grandmaster Kasparov would keep written files of all the important chess games he had ever played. This enabled him to go over past games finding out where he made mistakes and glean fresh insights from previous games.
As computers came to the fore, Kasparov started to use a portable computer. Instead of dragging around suitcases of written files, Kasparov could store all his games (as well as other players’ games) on his computer.
The ability of computers to store huge amounts of data means that an investor can more quickly accumulate the experience that comes from real-life investing. You can quickly go over past trades and historical events, you can replay trading sessions and you can practice trading even while the market is closed. Computers, therefore, can speed up the process of practice and trial and error.
#7. By improving feedback and analysis of investing decisions
Kasparov suggests that one of the main ways machines can help is by improving the feedback process. In chess, this involves the recording of games and post-game analysis. In investing, computer programs can be used to store trades (trades that are taken as well as rejected) and this data can be analysed at a later date to provide essential feedback.
You may think that a computer is not needed for this task and that you can get away with writing your trades down into a journal. But think of the increased power that a sophisticated program can provide.
A computer program can quickly analyse thousands of trades and present unique statistics such as risk:reward, maximum adverse excursion, or what time of day works best. Some excellent trade analysis services like TraderVue already exist. Or it may be a case of working on your own piece of software.
How investors can work better with computers
So far we have seen how computers can help investors to make better decisions. However, computers have major weaknesses so it’s crucial to think about what humans can do to work better with machines. Here are some important steps:
#1. By cross checking data
A computer can only perform as well as the data that is fed into it. As the old saying goes ‘garbage in, garbage out’ so human investors need to make sure to cross check data that a computer uses.
As an example, I recently ran a stock screen using data from Reuters and found that Apple was showing a market cap of $600 billion. I don’t know how this error happened or what caused it but this error flowed through to valuation metrics like price-to-earnings. If you had followed the computer’s lead here you would be acting on faulty data.
A computer has no idea that the data it’s using is wrong but an investor can get into trouble if she does not do regular cross checks. In this instance, a quick check of other sources would reveal that Apple did not have a market cap of $600 billion and thus the trading signal was in fact void.
This tends to be a bigger problem with intraday trading where data ‘spikes’ are more frequent. However, the problem exists everywhere. Investors need to be aware that when data contains errors algorithms are not reliable.
#2. By better understanding risk
Computers are able to calculate some types of risks and backtesting can give an idea of the comfort level associated with a particular investment strategy.
However, computers are not very good at estimating risk because their calculations are necessarily based on historical data. The future is not the same as the past so how can a computer respond to black swans or never seen before events?
In my experience, a skilled trader has a better intuitive understanding of risk than a computer model. This is because they have been through many different types of markets and know that the unexpected can happen at any time.
This is one area where human intuition can help an investor add value to his computer model. It is also why algorithms often blow up or are turned off during rare events.
Leaving risk decisions 100% to a computer model is a disaster waiting to happen as was seen with Long Term Capital Management.
#3. By gathering more information
It’s easier to overfit a computer model when it is made up of lots of different rules. A trading strategy with lots of different rules may find itself overfit to noise in the data and unlikely to work in real life. It’s this reason why computer models are generally kept quite simple – It’s easier to validate a simple trading strategy than it is a very complicated one.
However, simplicity has some downsides, especially in financial markets which are notoriously complex. A computer model, therefore, runs the risk that it is just too simple to work in complex, evolving markets.
A human investor can overcome this by gathering more information in order to supplement the trade. For example, most computer models are built using price and volume data. But there are many other sources of alpha that a human investor can consider.
For example, what is the price-to-earnings? What is the company’s management like? What is the current sentiment on social media? What trends are working in the company’s favour?
An investor might use a computer model as a base for filtering down investment opportunities then she might spend just as much time looking at fundamentals and qualitative factors before making an investment.
An important aspect of this approach is to make the gathering of information part of the process. There is no point spending time gathering information if it has no impact on the end result.
As Paul Slovic noted in his 1973 horse racing study, more information does not necessarily lead to better prediction. So it’s important to come up with a process that increases alpha without adding noise.
#4. By using intuition
I already said how experienced traders often have a better intuitive understanding of risk and intuition is a big area where investors can add value.
Pattern recognition is one of the main sources of intuition since humans have evolved to become supreme pattern recognisers and we are much better at recognising real patterns than algorithms.
However, intuition is a tricky thing to define since it is hard to measure and hard to explain. An investor often feels intuition about a certain stock but has no way of explaining why they feel that way. Often it is a result of learned patterns or stories buried in the subconscious.
Because of this, investors need to use intuition very carefully. Daniel Kahnemann suggests the first thing to do is to slow down and think through the intuitive feelings. You have felt something that may or may not be true. Now go and gather the facts to see if your intuition is correct.
Intuition can lead to the skipping of computer generated ideas. Or it can support them. Once again, we come back to that idea of process. Of continuously evaluating moments of intuition to see if they are adding value. If your intuition is not adding value then you may as well follow the computer, not your gut.
#5. By providing context
I have spoken of how computers have weaknesses and one of those weaknesses is that they operate in isolation without any context or concept of the bigger picture. Human investors are much better at this.
Michael Mauboussin gives the example of a fundamental analyst during the Subprime crisis – ‘a fundamental view can be of help—for example, if a model suggests to buy shares in a German bank when the fundamental analyst knows that German banks are heavily invested in subprime.’
In the past, some of my own investing strategies have recommended positions in fossil fuel companies as their share prices have sunk to oversold levels. However, I usually avoid them knowing that the world is transitioning to renewable energy.
Another example might be the influence of low interest rates. Such an environment provides a level of support for companies in debt so these stocks are likely to do better than they would in a more normal regime. As a human investor, you have the ability to step back and analyse whether a trading signal is in line with the bigger picture.
In fact, since algos are weak in this area, one idea is to seek out the mistakes that algorithms make and exploit them. For example, algorithms are able to read text based reports such as economic releases and earnings reports and trade almost instantaneously giving them a significant advantage.
But algorithms are typically scanning for certain types of words and they make mistakes. They sometimes buy the wrong tickers and they cannot possibly understand the full context behind a particular piece of news.
An algo might bid up a stock based on a strong earnings report but it might miss a crucial piece of information on an earnings call. It won’t happen all the time but a human mind is better able to evaluate the bigger picture and decide whether or not argos have made a mistake in their interpretation.
This is another example where a computer can be used to identify opportunities but a human is given the final say on whether or not to execute the trades.
Part of this circles back to what Buffett calls a ‘circle of competence’. You can only provide context to a trading signal if you actually have some expertise and insight to add.
For example, if a computer model identifies a trade in a small, biotech company I know nothing about, then there is very little insight I can add. Having said that, if the model is strong enough, any additional information you can find might be beneficial.
#6. By directing strategy and asking the right questions
One of the key areas where human chess players have historically had an advantage over computers is strategy. While a computer excels at running the numbers, a human player can use clever tactics such as playing slowly to draw an opponent in.
(In fact, this was one way that human players beat computers before they became more sophisticated. A human player would slowly move pieces around the board until a computer made a mistake and then the human player would pounce and quickly win the game).
In investing, it’s important to make strategic decisions such as what strategies to use, what sectors to focus on, how to allocate risk and how many positions to hold in a portfolio.
Backtesting is one way to get an idea of what strategies work and how they blend together. But it is up to the investor to come up with the initial ideas with which to test and a human is always going to be able to come up with more creative ideas than a computer.
Furthermore, a computer only provides answers to the questions it’s given. It is up to a human to guide the computer in a strategic, sensible way.
When I first began my investing journey I was skeptical of most quantitative methods. I believed that markets were too complex to be beaten with an algorithm.
However, my results revealed that I was not making particularly good decisions. My investing suffered from a lack of discipline and focus and this led to losses and missed opportunities.
I started to learn about quantitative trading and went deep into areas of backtesting and strategy development. This experience added necessary discipline and framework to my investing process that provided huge benefits.
As I progress further, I find myself leaning back towards the human elements that I used when I first started. I’m able to understand the strengths and weaknesses of both approaches and I find myself spending at least as much time analysing an investment from a fundamental standpoint as I do analysing it with a computer.
A key message that is clear from reading this article is the importance of building a process based on constant feedback. Using computers to form a smarter investment approach and to eliminate mistakes. All investors use computers to some degree but there is a great deal more that can be achieved, particularly as technology advances.