PETER BAKKER | Unhedged
PETER BAKKER | Unhedged
What is automated, algorithmic or black box trading? How is artificial intelligence used in stock picking? Are computers taking over from humans in selecting and trading stocks and shares. These tools are often used by hedge funds, but today we're going to be talking about being unhedged, a micro-investing app that allows you to invest algorithmically in the US market.
Peter Bakker is the Founder and CEO of unhedged. He's an algorithmic trading fanatic, seasoned entrepreneur and marketing executive. Peter has been trading using algorithmic methods for 12 years, with Zipline, Keras and other AI platforms. He has exited successfully from Track4 and other smaller ventures. As CMO, Peter led B2C marketing for Aussie Farmers. He also has an INSEAD MBA.
“If something accelerates to hard then it's more likely to go down. We use a very complex machine learning algorithm that detects bubbles and you can detect bubbles on daily graphs or hourly graphs or whatever. But the key is basically that you want to see an acceleration that is logarithmic: up or down. We're better at detecting the bottom than the top, because the top consists out of multiple bubbles in general. It's like champagne, there's bubbles everywhere. And bubbles not always explode together. They can pop one by one and then it's an orderly decline. If they start to correlate, like we had in 2008 especially, we correlated all the way down.”
Contrary to popular belief, you don't have to invest a lot of money to make progress when it comes to the market. Invest small & watch it grow over time! Unhedged lets you invest with as little as $100! You can add & withdraw anytime.
“The modern portfolio theory says that you have to do a lot of things. One of the things is you have to have diversification in your return streams. And what do they do then? They say, ‘Well, an ETF is a return stream. So, I diversify in ETFs.’ But the problem is that if markets go up, okay, it all goes up. But the moment goes up with the stairs, but it goes down with the elevator. The moment you go down, then things start to correlate. And then suddenly those diversification strategies don't work anymore. And so, that's where I think algorithmic trading is a little bit more pure to the modern portfolio theory because it really rebalances return streams.”
TRANSCRIPT FOLLOWS AFTER THIS BRIEF MESSAGE
It’s like having your own investment fund that uses algorithms to constantly scan, analyse and optimise your portfolio. The algorithms look at more than 5 million data points every single trading day!
Unhedged doesn’t know your personal circumstances. Before making a financial decision you should read the PDS and TMD and consider whether their product is right for you and whether you need some advice from a professional financial adviser.
G'day and welcome back to Shares for Beginners. I'm Phil Muscatello. Now, sometimes we hear about automated trading or algorithmic trading or black box trading, artificial intelligence kind of trading. This is where computers take over from humans in deciding what stocks and shares to buy and sell and how to buy and sell them. This area has traditionally been an area for hedge funds, but today we're going to be talking about being un-hedged. G'day Peter.
Thank you. Yes. Unhedged, correct.
Peter Becker runs Unhedged, which is an algorithmic trading platform, which is about less than a month old now, isn't it?
Peter (1m 3s):
Well, yeah. We started building the platform in December, 2020.
Phil (1m 8s):
Peter (1m 9s):
After we raised our first round. And so, we had to build basically everything ourselves, right? Because we can't rely on others to give us all the trading advantages. So, we built our own trading engines, our own algorithms and our own apps. So, yeah, in August we launched for wholesale investors and now in December 20th, four days before Christmas,
Phil (1m 35s):
Great time to start. Better late than never.
Peter (1m 39s):
We launched for retail investors. And so, now, because we had a wait list of about 8,000 people and we're still onboarding them and we release for the public in like two or three weeks, so, without any wait lists. Yeah. Really exciting.
Phil (1m 55s):
So, what you're trying to do is offer regular investors access to the kind of tools that hedge funds have.
Peter (2m 2s):
Phil (2m 2s):
What is a hedge fund?
Peter (2m 4s):
Oh, yeah that's a good question. A hedge fund is basically a class of funds that the regulator says "there are no rules" and it's only meant for sophisticated investors. And so, what we had to do is we took those tools from the rich boys and we adjusted them so, we can use them for everyday investors. So, that means that normally hedge funds will use complex derivatives. We can't use them. Or options for these, we can't use them. So, we try to create a fund that is clearly not a hedge fund, but that uses the tools that hedge funds use.
Peter (2m 50s):
Yeah. So, if you look at the hedge fund industry, if you want to know, there's a lot of funds that, for example, have a, what they call high frequency trading. So, they try to arbitrage like cents or millicents off each trade.
Phil (3m 7s):
They're just trying to shave off little bits, aren't they? Just by getting in a few milliseconds before the trade happens.
Peter (3m 14s):
Yeah. Or if somebody has a fat finger, then they will give it and then buy it back, right? So, they try to use inefficiencies in the market to make a profit. And then you have hedge funds like Bridgewater in the US of Dalio or double on capital. And they are more funds like we do. So, we try to find alternative data in the industry and then find trends before they're happening. And you can do it as in a lot of ways. And so, what we have chosen is the way that Ray Dalio's fund does as well.
Peter (3m 56s):
So, we use multiple algorithms that have total different data sets, rules, universes and when you combine them in a meaningful way, then it doesn't matter that one algorithm is too early or too late, right? So, as long as they don't make the same decisions at the same time. And so, they will basically make a equity curve, what they call more smoothly, aka your account.
Phil (4m 27s):
So, this is not your first foray into investing, you've invested before and you've failed spectacularly, I believe.
Peter (4m 35s):
Spectacularly a few times, a few times. So, if you go back like, maybe this says now that I'm an old man, but my first business, I started in ‘96. And it was a business in software engineering and primarily in solving something that in hindsight didn't exists, the Y K2 problem. And my business was a pretty successful and free minutes to sell it in 2001. And after working for five, six years as a dog, I decided to travel around the world and give my money to an asset manager who would invest in hedge funds and all professional and they convinces me it was great.
Peter (5m 24s):
And when I came back after two years, they professionally halved my money, which was a shock. But also like, what I thought was odd is that they didn't feel any regret. They said, "Yeah, that's the game." I said, "But I still paid my fees, right? I paid 2% a year over my assets." "Yeah. Yeah. But you know, that's it."
Phil (5m 47s):
And they didn't have perform an ETF at all.
Peter (5m 50s):
No, no, it didn't outperform anything.
Phil (5m 53s):
Peter (5m 53s):
And then I started to trade myself and if you start to trade yourself, like most people in your podcast, you try a lot of things. And so, I tried technical analysis, which in hindsight was actually not so, smart to try, because if it would work, then there wouldn't be a lot of more trillionaires in the world, right? So, technical analysis can work in some areas sometimes, but it's not a silver bullet. So, then I started to trade more aggressively in options and yeah, I did well, but the volatility of my account was just massive, right?
Peter (6m 34s):
So, I lost them a lot of cars in a day. And at one point I thought, "This is not a lie, it's bad for your heart." So, then I discovered algorithmic trading and it started basically in 2010, I started to write stuff. And first in Excel and then in code and Python and slowly, the world opened up in terms of APIs for brokers. And there was more data available. So, it became more and more sophisticated, I would say. So, then slowly over 10 years, I developed this theory also working with a lot of other hedge funds and smaller funds about that, a lot of robo investors, they abused, the modern portfolio theory.
Peter (7m 25s):
They basically, the modern portfolio theory is, it says that you have to do a lot of things. One of the things is you have to have diversification in your return streams. And what do they do then? They say, "Well, an ETF is a return stream. So, I diversify in ETFs." But the problem is that if markets goes up, okay, it all goes up. But the moment goes up with the stairs, but it goes down with the elevator. The moment you go down, then things start to correlate. And then suddenly those diversification strategies don't work anymore.
Peter (8m 6s):
And so, that's where I think algorithmic trading is a little bit more pure to the modern portfolio theory because it's really rebalances return streams.
Phil (8m 16s):
So, what does that mean? Rebalancing return streams.
Peter (8m 20s):
So, what you basically do, if you have algorithm one and algorithm two and algorithm, three, they all have a return, so, to give money back to you, right? So, you by and your sell. And when you sell, you have a return. They have their own volatility but the moment you combine them, they can cancel each other's volatility. So, you create algorithms that have so-called low correlation. So, that means that they behave differently.
Phil (8m 49s):
What's a practical example of that?
Peter (8m 51s):
Well, practical example of that is, for example, suppose you have a portfolio like Ark, yeah? Ark is this famous young lady
Phil (9m 4s):
Peter (9m 5s):
Yeah, Cathie Woods, who basically piles up and in Tesla, Apple, all the meme stocks, right? And so, her fund has a very specific return stream. It's very much leaning on the extreme edge or the, what in algo trading we call it, the momentum edge of the tech market. Now, if you put all your money into that, then you become basically an amplification of the NASDAQ. And the NASDAQ is already pretty volatile. So, amplification of that means that you go up more when the NASDAQ goes up and down more when the NASDAQ goes down.
Phil (9m 49s):
Which she's done recently, hasn't she?
Peter (9m 53s):
Well, it's, her fund has been obliterated, right? And so, if you do that, then that's not a good strategy. So, you can take a second strategy where you go to, for example, a company called PIMCO and PIMCO is famous for bonds strategies. So, they buy bonds in distress, they buy government bonds, all kinds of bonds, inflation protected bonds, they protect them for rising rates. So, although does is a very low returning strategy, if you combine the two, suddenly your portfolio goes up a lot smoother. And so, okay, you can combine like three, four or five.
Peter (10m 34s):
And so, the loss of diversification in your own portfolio is basically, if you have more than 15 to 20 stocks, you start to be over diversified. So, it's the same for us. If we have more than 50 algorithms, we start to be over diversified. So, over diversification is actually a cost to your portfolio. So, you don't want to do that. So, I always say to people who want to run their own investment that they have to limit the urge to buy a lot; a lot of types. If you see a portfolio with 200 things, how can you follow 200 companies?
Peter (11m 17s):
That's really impossible.
Phil (11m 18s):
So, diversification, you were talking about bonds. So, these are like corporate bonds, which is money that's loaned to companies
Peter (11m 27s):
Phil (11m 27s):
And then government bonds and so, forth. So, it sounds to me it's like diversification across different asset classes, is that correct?
Peter (11m 36s):
Well, yeah, that is one type. Then you have a lot of other diversifications which are more interesting. And that's what we algo nerds call factors. So, there's a whole bunch of factors we know. And the factors are basically data that explains a part of the returns of the market. So, if you look at the S and P 500 or the ASX 200, the whole market moves in a certain way, but sometimes the market is driven by what they call momentum. And sometimes the driven is by failure. And sometimes the value is a quality metric.
Peter (12m 19s):
And sometimes the market is driven by size. So, only big companies get bigger like you have in the US now. So, if you look at the current market in the US it was driven for a long time by size and momentum within in a bull trend. And that means that we have a momentum algorithm, we did very well, right? And so, in last December, when the NASDAQ went down, I think 7% or so, we went up 2%. And why? Because the algorithms already saw that the momentum was waning in the tech sector and it was moving to others. So, we had certainty a lot of utilities in there.
Peter (13m 4s):
We had Procter & Gamble and we had medical companies in there. So, these algorithms are far more tuned to the market and they can switch quicker than people because people get attached to an idea, right? So, Cathie Woods is still at the idea. If you look at Spaceship, robo advisors in Australia, they still very much on the tech trend, right? We're more agile
Phil (13m 31s):
Agnostic, agnostic to what you're going to be investing in.
Peter (13m 35s):
Another factor you can use is mean reversal. If things go too far off the mean, then they go back.
Phil (13m 40s):
That sounds a bit technical though, doesn't it? 'Cause you're looking at a chart and you're looking at the
Peter (13m 50s):
Phil (13m 50s):
At a moving average, aren't you?
Peter (13m 51s):
It's more, you mostly look at acceleration.
Phil (13m 54s):
Peter (13m 54s):
Yeah. So, if something accelerates to hard then it's more likely to go down. We use a very complex machine learning algorithm that detects bubbles and you can detect bubbles on daily graphs or hourly graphs or whatever. But the key is basically that you want to see an acceleration that is logarithmic: up or down. We're better at detecting the bottom than the top, because the top consists out of multiple bubbles in general.
Phil (14m 30s):
All bubbling up like champagne.
Peter (14m 32s):
Yeah, it's like champagne, there's bubbles everywhere. And bubbles not always explode together. They can pop one by one and then it's an orderly decline. But yeah, if they start to correlate, like we had in 2008, especially, we correlated all the way down, right?
Phil (14m 49s):
Everything did, yeah.
Peter (14m 50s):
And in 2000s was also we correlated all the way down. I think it helps for me, especially when I work with my quant team is our straightening into 2000. I knew what was happening, I saw it and I spoke to my dad and my dad said, "This doesn't make sense." And I said, "Dad," more or less, "Shut up, this time it's different." And how different was it? Not very. And my dad with his gold stocks was pretty right. Oh yeah, I wanted to go back quickly to how I failed, right? Because I failed a few times by giving money to others and then being not so, good at trading and then becoming better at trading.
Peter (15m 34s):
And then later, when I started to algo trade, I had to learn to trust the algorithms. One time in particular that I had made one algorithm was really good at detecting stress in a market. And a few times it took me out really quickly and I thought, "Oh, that's really good." So, I made a lot of profit from that algorithm and went from possibly 15K to 150K in a few years' time, like bizarre. An I thought "It's really
Phil (16m 7s):
This is the magic one.
Peter (16m 8s):
This is the magic one, this is the boat, he's coming." And then one night I was sitting and I thought "That's really odd that's he jumps out." So, I overrode it. And I thought "That's really odd that the stock goes in 2018 when the volatility crash had happened." And so, during the night I spent overriding the algorithm who wanted to jump out all the time and acquiring more and more of something that started tonight at $110, I think. And it, at the end of week was $4. I lost $80,000. That's my big
Phil (16m 45s):
By not listening to the algorithm.
Peter (16m 48s):
By not listening to the algorithm. This is the last time I overrode it an algorithm because the algorithms, you train them on data. You retrain them on data. And at one point they find a pattern and the pattern might be way too complex. And how you can explain it, I think, is people are linear people. We are linear thinkers. We think that, you know, COVID is going up. Oh no, go up forever. Or have in markets go down and people predict everywhere. Oh
Phil (17m 20s):
They'll always go down.
Peter (17m 22s):
Yeah, we'll go 70% down because we're already 30% down. So, people are linear. But most processes in the economy and in the market are non-linear. And we just don't have the brain to understand 5 dimensional, 10 dimensional, 40 dimensional plots and computers are. Computers are very bad at explaining. They just say yes or no, right?
Phil (17m 53s):
Peter (17m 53s):
Or on or off or with a confidence of 15% off or confidence of 20% on. But they can't say why. It's really hard. And that's something you have to get used to when you're an algo investor, is that sometimes you look at the market and you think, "Hmm, I don't know why he does that, but I just let him do it." And as long as you play with different mechanisms, different factors, so, momentum, mean reversion, size, quality, maybe earnings, it can also be a factor, then you'll be fine. I think is more scientific way of investing.
Peter (18m 34s):
And I must say that during the time that I do algo trading, I never had a drawdown on the algorithms more than about 9%. Which is really good if you see that the market window, 35%, right? And, even, yeah, like December, I founded 2% plus. Well, I know because we opened accounts and other competitors, they went down by 6% because yeah, they are very married to this idea and the algo's are far more agile.
Phil (19m 6s):
With algorithmic trading, an important aspect, I believe, is back-testing. Do you do back-testing? Is that how these algorithms are developed to see how it's worked in the past?
Peter (19m 17s):
Yeah, so, back-testing is a gift and a curse at the same time.
Phil (19m 21s):
So, this is when you apply the rules that you've come up with the algorithmic rules and then look at it over the last 10, 20 years and see how it actually works.
Peter (19m 30s):
Yeah. Yeah. I'll explain it a little bit further. A back-test is basically, you first start with acquiring data, right? When you're an algo trader you acquire data, so, you have found a data set that's really interesting that there's some alpha in there or there's some factor in there that you want to abuse. And you split the data in a training set and a validation set and then you have the forwards trading set. So, these things can be sequential or non-sequential.
Peter (20m 10s):
So, you can test with the data 2012 to 2015 and then tests back. So, test in 2010 or test forward. So, you can test a lot of ways to validate it. The problem is this it's been academically proven, people who do a lot of back-tests, they're curve fitting. And they basically try to find a way to follow the curve.
Phil (20m 33s):
And that curve is the price? Price of the markets
Peter (20m 37s):
You say, "Well, every time in the market does this, I do that. And therefore, I make more money." But Fukushima only happened once. COVID, now, only happens once. And there's a lot of things in the world only happened once. So, once you start to curve fit your algorithm to those events, then suddenly you are, what they call, curve fitting. So, you're lying. And a lot of amateur algo traders are doing that. And you'll see that also in all the blogs about technical trading often is that people say, "Oh, this always works." Well, when you put it in a common computer, it doesn't.
Peter (21m 19s):
But yeah, it's a good try and it works sometimes. And you have to just know when it works and when it doesn't work. The back-testing is a fantastic tool to see what would happen in the rules you have developed in your AI model. However, when you back-test, you have to take into account, for example, that you have no what they call information leakage. So, for example, if you back-test in sets and you basically look back to data that you actually developed on, then that' what they call information leakage.
Peter (22m 1s):
So basically, your model is actually lying to the data in front. And so, there's a lot of things you have to be careful with. The second thing is that when you are trading in the markets, right? You get the data stream as it presents at that moment. That data has misprints, that data has outages. You know, Amazon can go down for a few seconds. The internet can be disruptive. It can be where your server is located. It can be a power failure. And this data that comes in your algorithm day by day is dirty.
Peter (22m 43s):
The data you use to back-test is clean. All these errors are taken out. They call it misprints. But misprints are still in daily life. So, a lot of people, when they do it back-test, they do a back-test on clean data. And especially in the crypto sphere, that is really, really important. In crypto you have thousands of exchanges and they all have their own price discovery. But when people do algo trading on crypto, they take the whole market and they put it in one dataset and you can find fantastic things. I can make albums on Bitcoin that always work. The problem is you can never trade it because you have to train it on exchanges and the parcels you can buy and settle on the exchanges are just too small.
Peter (23m 32s):
And the friction, the buy sell spreads and transaction costs are just, in the end, killing you. And that's the last thing I would say with back-test is there's a lot of people would do back-tests for fun. You know, they forget to calculate slippage and slippage is
Phil (23m 48s):
The brokerage, is it?
Peter (23m 50s):
Yeah, it's brokerage, but also the buy sell spreads, right? So, you have to brokerage costs, which you can calculate. But you also have what they call slippages. For example, if the order book of the exchanges has a thousand shares on a thousand dollars and another guy who wants to sell a thousand shares for $1,010, if you do now in order for 2000 shares, your average price will be 1,005.
Phil (24m 23s):
Peter (24m 24s):
Because you buy both parcels because the moment you bought, it's depleted and then it takes up. That is very hard to mimic. And we spend an awful amount of time building software that actually uses that book to read it and say, "Okay, a parcel should not be decisive." And that's the power of algorithmic investing. We can read life. Gigabytes of data interpreted and get the best outcome for the client.
Phil (24m 54s):
So, tell us about Unhedged.
Peter (24m 55s):
Phil (24m 55s):
How can people find Unhedged and it's an app now, isn't it?
Peter (24m 58s):
Yeah, it's an app. So, you can download it. If you download it now, you have a demo mode, so, you can see what it is and in about two weeks we will open it up for everybody. We're still onboarding our waitlist but that will be quickly done because now we have the confidence we can onboard a few hundred people a day. But yeah, Unhedged, go to the Apple store or the Android store or go to the website unhedged.com.au. And that is U-N hedged.
Peter (25m 39s):
Phil (25m 41s):
Peter (25m 42s):
Yeah. I think it's my accent that makes unhedged, unhinged sometimes. And so, just to be clear, Unhedged in the background is fully app-based and it has a fund in the background. It's a unit trust so, it's totally regulated.
Phil (25m 58s):
You're buying units in this, yeah.
Peter (25m 59s):
Phil (25m 59s):
What sort of investment can you start with?
Peter (26m 3s):
So, the first investment is from a hundred dollars confidence, so,
Phil (26m 7s):
Peter (26m 7s):
You got to start with almost nothing. If you want to invest more than 10 million, you have to call because then... Nah, but we're gaining funds on a management really quickly, which is for us a great validation. Because, you know, in the end you can make up everything and you can build the best team there is, I think we have an awesome team, but in the end, people fall with their wallets. And if we see that people now just put 50,000, 5,000, 10,000, 15,000 easily in, I think, yeah, it's really resonating.
Peter (26m 48s):
It's really resonating that we try to make the investment world a little bit more fair. We're giving the tools of the big boys to just people like you and me.
Phil (26m 58s):
Average investors, yeah. Okay. Peter Bakker, thank you very much for joining me today.
Peter (27m 2s):
Shares for Beginners is for information and educational purposes only. It isn’t financial advice, and you shouldn’t buy or sell any investments based on what you’ve heard here. Any opinion or commentary is the view of the speaker only not Shares for Beginners. This podcast doesn’t replace professional advice regarding your personal financial needs, circumstances or current situation