Artificial Intelligence and Trading and Investing
Exchanging Ideas — Artificial intelligence in trading | London Stock Exchange GroupTo see the full interview at the London Stock Exchange see this link above.
Look at the image below.
The red arrow is generated by our trading algorithms, telling us to ‘short’ or sell the market, in the expectation of profiting by buying back later at a lower price. ie buy low, sell high. Or in this case sell high, buy low.
As computing power has evolved algorithms have improved to try to forecast market direction. We are trend followers, and look for trends based on momentum of price moves and patter recognition.
‘AI’ or neural networks as a subset have existed for a long time and not all processes are as good. Often times by learning from the rules of the game, they’ve so often ‘curve fitted’ so only worked historically. Trading as a game is not like chess, where you can tell a machine the rules and it works out the optimal moves. The other way chess machines learn, by experimenting, and seeing outcomes and probabilities, is more like trading. And this is the type of ‘AI’ learning we prefer.
Pure pattern recognition alone is not enough either for an edge in trading. We need more than a 50–50 coin toss edge of a move to have a good trade worth taking. Pattern recognition alone, does not give that. The old days of Japanese Candlesticks, wedges, flags, pennants, work. Yes, they all work. Just not often enough. A little like playing the lottery, you win small prizes often enough to keep going. But nothing that would justify leaving the day job or even a net positive return over the long term to make a business out of it.
We found out ‘edge’ therefore by examining more than just patterns alone. We also examine the price movements, speed, and the moves inside the time frame and across time frames. All of these have a logic to them as to why they should suggest a probably move in one direction over another.
Machine learning techniques such as regression analysis in trading have been around for a long time. And again, rarely effective enough to work long term — the markets are just too open ended — what can happen is too variable for this type of analysis.
Trading is more suited to machine learning techniques of anomaly detection, clustering and classification.
Return to our image again. Notice how we did not pick the exact top. Notice too the move did not happen immediately following the signal. That is characteristic of ‘clustering’ which is more likely to succeed in the markets than is regression.
The promise of deep learning is more accurate forecasting. We are as confident of that as the universal theory of everything in physics!
To learn more visit www.trading-champions.com
Trading Algorithms for the Masses? The Google of Trading
I have always been suspicious of social trading. The idea that some IT engineer sitting in their underpants in St Petersburg is going to make a consistently profitable trading algorithm. The reason is not that their computer skills are weak.
It is that necessarily in trading there are losing trades, there are unrealistic expectations that computers mean only winning trades, and that drawdowns don’t happen.
When I have looked at algorithm performance on a multitude of social trading sites, their performance does not come close to what a hedge fund like ours could ever raise money on from institutions. And private investors tiny capital isn’t going to make anyone money — unless as many of these brokers offering social trading do, you gouge the clients trading with hidden costs. Not exactly a good business model.
Or you raise a bunch of money from VCs on hype and then the VC and you sell to the bigger fool. What you might call the ‘Silicon Valley’ business model. But what about a Google type long term sustainable business model — something that is profitable enough to sustain billion dollar fines and not blink?
I’ve always dreamt of being on a boat in some warm climate, and my little server enters and exits the market. But I knew I could never spend millions needed for the HFT type of trading. So how could this be done?
We took two approaches. First, signals software where we would place the trades ourselves and us having ultimate control. But those signals based on scanning of patterns across multiple time frames and currencies.
The second approach, an auto-trading robot on our VPS which can execute trades for the client through their own broker account on the free ubiquitous MT4 platform. But allow us to monitor it 24/7.
Why offer it to the masses? Profits = buy price — sell price. This broad rule seems to miss a lot of people. A string of code unless sold is worthless. Coca-cola’s secret formula still needs them to sell the code in every single tin and bottle and glass.
But how is that ever worth it? For us, we took a dual approach again. First, continue offering institutions and HNWs trading. The old way. The old way which may be under threat. But we also realised that technology companies become billion dollar companies long before hedge funds do. Maybe that’s why Jeff Bezos left being a hedge fund to set up Amazon.
We don’t want to charge anything for fund management. We don’t want to be charging 2/20 — it’s dying. Instead, they will pay like they do to Microsoft — the SAAS model of Fund Management — a fee to use our VPS to place their trades or get our signals. No entry, exit charges. 24/7 access. And no need for us to be strangled by regulation. The broker has to be regulated and do the KYC.
Death of the Office: We keep hearing the expensive Mayfair offices and the 9–5 is dead. Well under our model — it is. Good. No more expensive rent. Fintech unlike hedge funds doesn’t need those expenses. A garage will do!
Model 1: Our Signals, but Your Choice looks like this:
But because you have the trader placing the trades, you give them mentoring and support. But under the subscription model, $2k pa for the signals and $2k pa for the mentoring, a global audience, different products, and it is not long before you are looking at a valuation for a tech company in the hundreds of millions.
Model 2: Our Signals, Placing and Cancelling Trades Automatically on Your Broker Account:
This places and cancels dozens of trades across multiple products and only about 1–2 get executed daily. The strategy has several substrategies. It uses quants, machine learning, artificial intelligence to pattern spot and adapts as the markets adapt and change. It uses technology already freely available like Uber uses cars and Google maps we use MT4 and VPS.
This is not HFT. It is not signals service. This is the new way algos will trade for the masses. The middle ground. And it works.
This is the winning formula of trading for the masses. Not social trading, not DIY brokers, though they want to work with us and we can, not algo developer sites, not fund management, not HFT.
For me, the first best trader I ever met, who I looked upto as a mentor was former Global Head of Forex at Salomon Brothers, Bill Lipschutz. He was in my book, the Mind of a Trader and in 2016, Hedge Fund of the Year. My ambition: for the apprentice, to become the master.
Why is this needed for Private Investors?
Now that brokers have to confess just how few of their clients make money (it’s not 49%) it’s 15% often, trading is truly a dirty word.
Take this broker tweet:
I am glad they have to disclose it. But the problem remains:
- brokers use in house ‘educators’ who are not professional regulated traders and so brokers continue to be dishonest
- brokers once they game the client acquisition entry form to get the client in (except a few they must show the regulator they rejected) they do not monitor to the size and speed of losses — a key indicator of unsuitability
That’s a shame given my second book was published by the Financial Times and called ‘Trading Online’ but disreputable and even those claiming to be reputable brokers have killed the word and the market. Like journalists, the majority have spoiled it for the few good ones.
Our hedge fund trades. Like most hedge funds. We have Sovereign Wealth Fund as an investor as well as the usual pension funds, family offices and high net worths. We are reputable and regulated and of course do not have retail clients.
Or take trading by the likes of Winton Capital. This slide explains in simple terms what a reputable respected hedge fund like them does in trading.
Given our style of trading is not the hyper fast, but more like in the image below where our algorithms tell us such trends, I propose to call it Fast Investing rather than trading.
Why not just buy and hold investing like Buffett? Well, we like to profit from falling markets too for one thing.
But the bigger point is trading has begun to mean mindless churning. It’s not what professionals do. Investing has, thanks to Buffett, a good name. We are at the trading end, but we are Fast Investing! And in a volatile world of tweets moving markets let alone Brexit — it’s a good job too not to be locked in too long.
Can the Private Investor Replace Be Their Own Fund Manager?
Well either fund managers will get better with AI, or private investors will be their own fund managers.
The reason, they are so poor at their job — they’re a marketing outfit, bringing in capital that is so hard to deploy that they index track anyway. So why do they exist…marketing spend. Thankfully entrepreneurs and companies in which they invest produce returns (often) and cover up for their poor performance compared to index trackers.
The age of the anti-expert and the empowered angry citizen is the perfect time to learn for yourself.
The one good thing fund managers do, lock up your money and stop you spending it or gambling it trying to trade without knowing what you’re doing. So they have some purpose. It’s just you could do better. Robo Advisors are potentially better, but only marginally. They clump you up into usually 3 baskets — low, medium, high risk. Hmmm…and again, charge you for picking some Exchange Traded Funds.
It’s a conspiracy of the old, tired ways, getting the money thank to inertia and marketing spend.
Worryingly Fund Managers seem to pick admired companies.
Source: Stocks of Admired Companies and Spurned Ones, Deniz Anginer and Meir Statman.
There is good news for fund managers…you do better after you’re fired by your employer…which helps you get re-hired…and the cycle goes on.
Source: Amit Goyal and Sunil Wahal, “The Selection and Termination of Investment Management Firms by Plan Sponsors” Journal of Finance, Vol LXVIII, №4
But Can the Private Investor Do It For Themselves?
A Process for Mining For Wealth Creating Stocks Yourself
Outstanding Investing Results: AI, Machine Learning, Data Mining, Algos?
For our performance, it was not really any of the above buzzwords. Yes we use computers to datamine. But data mining can lead to curve fitting. Yes, artificial intelligence sounds clever, like neural networks did — but lead people to forget that the world is not foretold like in a crystal ball.
It is not about winning across every trade. It is understanding that as a collective the performance has to be consistent.
Back-testing is useless and not something you can market an asset management company on. Why is it useless — lack of credibility. It may help you narrow down your focus area, but beyond that — you cannot use it for marketing or gaining assets.
When our software launched in 2004, we knew only time will be the best judge. Not clever buzzwords looking for the holy grail.
Just as Big Blue, the IBM supercomputer beat Gary Kasparov the world chess champion, so our Big Blue, ‘APSE’ has defeated Warren Buffet, the world investing champion and richest man in the world.