Do these mistakes put you off?
Monday, 15 April 2019
By Jason McIntosh
- A potent combo
- Be honest…
Algorithms aren’t perfect…
Yes, you read that correctly. I, the algorithmic trader, saying algorithms aren’t perfect.
Just like you and me, algorithms get things wrong.
GPS devices are a classic example. They mostly direct us with ease and efficiency. But occasionally they’ll get ‘lost’ and require a human to sort things out.
Take this for instance…
A trio of Japanese tourists got a lesson in algorithmic errors in 2012. The students set off on a daytrip to Queensland’s North Stradbroke Island. But it all came to an abrupt end.
You see, their GPS didn’t account for the water crossing. The tourists got 500 metres into Morton Bay before getting stuck. They say the device was directing them to a road.
Laugh, you may.
But this story isn’t a one-off. You’ll find many instances of GPS devices getting it wrong.
And then there are driverless cars…
The algorithms behind these vehicles are amazing. But just like their GPS cousins, they aren’t infallible. There’s already been several fatalities, along with a string of lesser incidents.
So should you be wary of algorithms?
In a moment, I’m going to show you mistakes from my own trading algorithms. You’ll then be able to weigh these against the potential benefits.
I’m going to tell you about some research I recently saw.
But first, a look at the market…
Over the weekend, the Dow Jones Industrial Average closed up 269.25 points, or 1.03%.
The S&P 500 gained 19.09 points, or 0.66%.
In Europe the Euro Stoxx 50 index finished up 12.49 points, or 0.36%. Meanwhile, the FTSE 100 rose 0.26%, and Germany’s DAX closed up 64.73 points, or 0.54%.
In Asian markets, Japan’s Nikkei 225 is up 323.64 points, or 1.48%. China’s CSI 300 is up 1.16%.
In Australia, the S&P/ASX 200 is down 4.30 points, or 0.069%.
On the commodities markets, West Texas Intermediate crude oil is US$63.62 per barrel. Brent crude is US$71.39 per barrel.
Turning to gold, the yellow metal is trading for US$1,290.30 (AU$1,799.58) per troy ounce. Silver is US$15.05 (AU$2098) per troy ounce.
One bitcoin is worth US$5,178.89.
The Aussie dollar is worth 71.72 US cents.
A potent combo
Imagine the setting…
You’re the chief admissions officer for a university’s MBA program.
Demand for the course is high — there’ll be many who miss out.
Your job is to select applicants who’ll likely be most successful after graduation. Get it right and you’ll receive a financial reward. Get in wrong too often and you’ll lose your job.
There are two ways of making your selections:
- Go through each application and handpick the students; or
- Use an algorithm to make forecasts based on past student intakes.
Which would you use?
Remember, there’s a lot riding on this. Do you put your job in the hands of an algorithm, or should you back yourself to make the best decisions?
This was the choice researchers gave participants in a recent study. It probably comes as no surprise that most people backed themselves.
But this next bit may surprise you.
People continued backing themselves even when they knew the algorithm had greater accuracy.
Researchers at the University of Pennsylvania (Dietvorst et al. 2014) found that many people opt for human forecasts, despite knowing an algorithm is better. This is in line with other studies.
The researchers call this ‘algorithmic aversion’.
So what causes this?
Well, it all comes down to confidence.
People are quick to abandon an algorithm after seeing it make a mistake — even though the overall outcome may be better. They’re also more forgiving of their own shortcomings.
You can read the research paper here.
Another interesting study involves the game of chess.
The highest ever rating for a human is 2,882. This was by world champion Magnus Carlsen in 2014. By comparison, today’s supercomputers have a rating of around 3,300.
A victory for algorithms, you may think.
But not so fast.
According to Tyler Cowen in Average Is Over, the best players aren’t human or machine — it’s a pairing of the two. You’ll see this happen at freestyle championships.
So does this make a chess grandmaster with a computer unbeatable?
Surprisingly, it doesn’t. Cowen notes some big failures by grandmaster/computer pairings. Their downfall is often due to an overconfident human ignoring the computer.
It turns out that the top teams are often a computer and a strong club player.
Chess isn’t the only area you find algorithm/human pairings…
An example you may be familiar with is Quant Trader. The algorithm identifies trades, and members decide which ones to take. Just as in chess, this can potentially be a potent combination.
So how does algorithmic aversion relate to trading?
Well, the previous study notes unrealistic expectations. It says that many people were expecting the algorithm to be perfect, and were put off when it wasn’t.
I believe this mindset makes algorithmic trading difficult.
Have a look at this:
This is a recent algorithmic ‘error’ from my own portfolio.
My algorithms gave a buy signal for Reginal Express Ltd [ASX:REX] at precisely the wrong time. The system’s entry point was at a peak in the share price. The stock fell heavily soon after.
Many people will jump on this and say the algorithm got it wrong. And they’ll ignore future trade signals. The algorithm, they believe, clearly doesn’t work.
Then there are examples like this:
This was a trade I took in Appen Ltd [ASX:APX].
Unlike REX, my algorithms got the entry right. I made a 115% profit within 13 months.
The problem this time was with the exit.
You see, my algorithms calculate the exit point for all of my trades. This removes any guesswork about when to sell. It also helps me manage my trades consistently.
But it’s not an exact science.
You’ll see that APX hit its exit stop on the day it made a lasting low. The shares then rebounded and haven’t looked back. If only the algorithm had set the exit point a few cents lower.
Situations like these make some people wary of algorithms. As the researchers found, many people focus on what an algorithm gets wrong — not what it gets right.
Check this out:
This is another trade from my portfolio — a trade in Jumbo Interactive Ltd [ASX:JIN].
Now, I’m going to ask you a question…and you need to be honest with yourself.
Could you get this outcome without an algorithm?
Here are some thinking points (align the numbers with those on the chart):
- Would you have held on during a 9% fall over just three days?
- Would you have resisted taking a 45% gain in under five months?
- Would you have held your nerve during the 22% correction in mid-2018?
- Would you have resisted banking a 100% gain within 10 months?
- Would you have been patient during 16 weeks of sideways trading?
- Would you have had a plan let the trade run by 504% (and still counting)?
- Would you know your exit point the entire time?
- Would you have even heard of JIN in the first place?
Yes, an algorithm can get things wrong. But it can also get a lot right.
The researchers say this:
‘Many decisions require a forecast, and algorithms are almost always better forecasters than humans.’
I know that my own trading is better with an algorithm. My decisions are more consistent, my use of time is more efficient, I find more opportunities, I have less stress, and I make more money.
This could be a reality for you as well.
I believe your advantage is that you know algorithms better than most. This gives you the opportunity to trade like few humans can on their own.
The age of the algorithm is here. I encourage you to use it to the fullest.
All the best,
PS: Humans come with a range of emotional biases. This can make for an entertaining afternoon supporting your footy team. But it can wreak havoc with your share portfolio. If you want to learn how many of the best traders operate, then sign up for Jason’s upcoming Trader’s Workshop. Its aim is to improve your skills and expand your opportunities. And the best bit is that it’s free for Port Phillip Publishing subscribers. Don’t miss this opportunity…sign up today.
PPS: All graphs and images above sourced by Author, unless otherwise stated.