002 - Validate for Yourself! The Science of Trend Following - Richard Brennan

Welcome to the algorithmic advantage we're here to expand the toolkit of the Quant trading community and introduce investors to the many advantages of systematic trading our goal is to educate and Inspire as we embark on a captivating journey into the vast knowledge and experience of leading portfolio managers and other experts in the field we hope you enjoy the show and if you do please subscribe leave us a Review or even buy us a copy by the link on the algorithmic advantage.com we really appreciate it [Music] welcome back everyone to the algorithmic advantage here we are

on episode number two where rich and I are just speaking amongst ourselves still before we go out into the big bad world and interview the other um great uh portfolio managers and and so on that we have lined up So you're stuck with the two of us for one more show and you get to get to know us and how we trade a little more so today I'm going to be asking Rich how he trades and um and what his uh strategies are all about so let's launch straight into it rich how you doing welcome

Simon very good thanks yeah I'm good um well mate I think as always we will uh start with a little brief uh bio a little brief background on yourself and Um how you got into the markets and and ultimately how you got into Trading sure Simon so uh back in the the old days of 1985 came out of college a couple of degrees under my belt thought I knew everything um went into Finance got involved in um the the financial controlling operations of a listed company and in that environment the enlisted company changed its course

um into the financial services industry And I therefore picked up a few extra hats in my role in financial controlling in areas such as compliance fund Administration fund management working for what they call a responsible entity which is a trustee here in Australia um it was a great opportunity to get thrown into the deep end and learn extensively about risk particularly in relation to risk that The controlling entity had in relation to the investment managers have plugged into its Network and I was also exposed to a lot of the the trading and investment processes adopted

by the investment managers so that led me initially in the path of of fundamental or value investing and that evolved over time to where I am now at the same time there was a strong passion I had with science and I had these two strands of my career path in Finance and my my science Journey they were traveling in tandem together and the the the scientific path forced me to challenge a lot of Notions that were found in economics with scientific principles and that led me into a fairly counter-intuitive space now where I'm now what

you call a diversified systematic Trend follower did um when you say that interested in science I know you're very interested in in physics for example was that just a Personal interest or had you was that part of your studies it was um my my courses at University were science-based and when I came out um into the financial world I always wanted to assess things economically in terms of scientifically so um I got a huge slap in the face from my father who I admired and respected amazingly so when I came out of college I thought

I knew everything and I went To my father pronouncing with confidence that I I knew a lot and he uh he in his very elegant way put me down and told me how I perhaps might not know as much as I think I know and to go along a path of what he called Step um standing on the Giants at the shoulders of giants standing on the shoulders of the Giants those who have investigated these queries that I was thought that I knew the answers of and in that investigation in my science path it it

demonstrated How little I knew and over the course of time the the investigation into science enabled me to challenge traditional economic theory and look at different models that perhaps were more applicable to more real-world situations than what was being announced by the economic community so the the science path to me was sort of a bit ahead of the economics path and so that gave me a good heads up into looking at these Opportunities that were perhaps a bit more advanced in the scientific world in the physics world and then applying that back to the economics

framework so you know over the last 30 40 years the the economics framework up until about the 1980s was severely entrenched in this concept of the official markets hypothesis and in the 1980s there was a slight evolution of new ideas new models new theories that were really challenging Concept and they were typically coming from the scientific background so that's where I got very involved in in a multi-disciplinary approach to trying to assess these markets which I view as complex adaptive systems and from that approach I think I've gained insights that have really helped me establish

the niche that I want to occupy in trading and investing these markets so how early on did you get involved in in actually starting to test Something or test strategies and get into that systematic trading stuff and you know when did software become available or were you doing things in Excel like how did that how did that originate when I started out I I was always deeply involved in systematic processes empirical validation that you know I I was trying to adopt the scientific method for economics so uh the way I do that is I'd have

a model of how I think a system Behaves and then I would validate that through an empirical process typically through a form of back testing process so back in the day we didn't have necessarily the computer power that we have so I was using things like the Huntley reports which were written reports of looking at the fundamental value of different companies price earnings ratios all of these things all sort of in text text in a big book that came out monthly the hunt the Huntley Report and then I was transferring that data into an Excel

spreadsheet and then evaluating the my intrinsic value of these particular companies and so that that's how I started out but then as as technology grew and grew this became far easier and therefore I started looking at systematic ways to really assist my efforts and Fast Track the the analysis to a point where I'm very comfortable now in the speed that I can assess and validate things using fairly rigorous Analytical computer assisted methods great all right so that's fascinating um if we were to move along then into I guess what frames that trading style that you

have now what's your broader philosophical approach um to the markets coming out of that science and um uh systematic background so I suppose the the broad frame of my Approach is that I view these markets as A very competitive landscape and in that landscape I view the participants in that market so I'm very much looking at what we call agent-based models the uh the behavior of an individual participant and the behavior of collectives how they all interact together and how that moves price so the way I view these markets in a very competitive landscape is

I think that the majority of the participants in that market are what I call predictive trade It's predictive investors they're using principles of extracting an edge through pattern recognition predictive processes that basically say based on my assessment of a back test and a rigorous application of that back test my assessment of the patterns in that back test or the data in that back test I with a fairly High degree of confidence can project that into an uncertain future with a degree of reliability and so I looked at that and I saw that about 90 of

the market participants were doing that it was not just industry itself that was doing that with their particular models of of theory but it was also um you know when you go to the YouTube blogs and you look at um you know these gurus on YouTube how they tell you how easy it is ETC they're all umbrella within this predictive mindset effectively which I think is about 90 of the market now I recognize That there is a degree of predictability in the market and there is an edge to be extracted from that but it's an

incredibly competitive landscape 90 so the edge has continually eaten away as uh more sort of uh you know the Renaissance Technologies of medallions Etc you know they get most of that edge the stragglers with less resources that the space or get less of an edge some of them are lucky and get a bit of an edge or whatever so that 90 is that area so I Said to myself let's don't go down that path because in my scientific background I knew that these there are these features of complex adaptive systems which were inherently uncertain so

um this meant that if you look at your your different systems that you can deploy in the market there are sort of four broad categories there's the simple system which is the what I call the no knowns the pendulum a predictable pattern oscillating two and fourth That's a simple system you get a complicated system which is what I call the known unknowns this is where you know the underlying Theory you're just presented with a novel system and that underlying theory is sufficient to explain that theory the known unknowns that's a second class of system there's

a third class system which I call complex and that complex system is what I call the unknown unknowns because there's a level of such high complexity In that system you don't know whether it is inherently indeterministic or simply a predictable model but an incredibly complex predictable model so this is where we had this unknown unknowns and in that third class of models there's a split there's that which falls into the predictive class and there's that that falls into the indeterministic class of what I call the fourth model which is the chaotic model and the chaotic

model um I I believe to be fundamentally Unknowable no matter what technology you throw at it there is a third class of indeterministic system which we see in certain complex systems such as the weather such as I believe the financial markets I believe that um understanding chaos is not simply assuming it's chaotic all the time that that's not her at all what chaos does is it has protracted periods of what is perceived to be predictable conditions punctuated by what I call endogenous Events arising from within the market at structure itself so this is different to

say a news Trader who is looking at external events that Force what I call forcings into that system they're from an external origin coming to force a change in that system there's this inherent endogenous complexity within the system itself which is sufficient to create chaotic indeterministic systems so my knowledge of science therefore said right that's a very complex world And that's a world where I can out-compete Ai and I I can out-compete the predictive Trader because this is inherently indeterministic how do I adopt models how do you trade something that sounds as complex as you've

just described in as unpredictable and noisy yes so the way I backtrack is through simplicity so um these incredibly complex models and you Might have seen these examples such as the double pendulum the triple pendulum where a single pendulum is a very simple yes a double pendulum and a triple pendulum creates a much more chaotic pattern now when you get this chaotic I have seen those so I think you can look them up on YouTube I suppose can't you reach where you see people swing a pendulum that might have like an elbow in the middle

of the arm and suddenly the behavior is Dramatically different and unpredictable the the what it will Trace out as it were is suddenly uh that that's exactly right so this complexity in the in the trace that it makes makes predictive models incredibly hard to navigate that complex environment but um someone like me will attack this problem saying hey don't worry about predicting it look at a risk mitigation system a system you can deploy in that Complex environment that is very simple in nature but allows for this complex trajectories and this is where I come to

the the age-old wisdom of the trend following mob the cut loss of Short net profits run you know in the is in these systems which are inherently chaos chaotic I also know from science that there is another feature of these systems they are what I call asymmetrical in introjectory what that means is that They are in a complex physics term it's called time asymmetric you can't reverse them so when you get to complex systems in science you deal with this concept called entropy and entropy is a a very very important thing when examining complex systems

because it's saying that the system is is has an arrow of time to it there is a forward motion of that complex motion that forward motion means that an air you can watch an egg shatter into a million bits But if you see that shattered egg in a million bits come back to being an egg you'll know that something is wrong with that video you know that it is backwards in time and this is because when we look at entropic systems such as all living systems such as the galaxies such as what what's driving this

universe um you see that it is a a condition driven by entropy a lot of people think it's a condition driven by energy but um it's a condition driven by entropy And what what entropy does is it says Hey in all this chaos There are rules being applied here not only at the small level but at the big level there is rules of structure which are dynamically interacting with the small and the large and this is where in science you get these complex models such as general relativity to explain the large-scale structure of the cosmos

which is about how the small interacts with the large how Matter is told how to move by space time and space-time tells matter how to move there's this Dynamic relationship about the two this is what I see happening in these complex adaptive systems they're in this mad frenetic complexity there is what I call fractal structure and in that fractal structure it is being driven by large structure and its interaction with small structure at all scales and in this world of the fourth box of Chaos this is what is driving um that uncertainty uh that is

the area that I want to get exposed to because I know that 90 of the market out there don't want to play in that game that's that's a game of chaos that's a game of uncertainty what they'll think for themselves how do I deal with that level of complexity with my predictive models and this is where I say well with the little trick up Our Sleeve with our simple models Um of bringing that you know fascinating uh philosophical approach about you know the reality of just how complex the the markets are and the difficulty then

of analyzing them what's your give us a little more detail on your your strategy so you said you've taken a trend following approach what kind of markets do you trade um over what kind of time frame and um and speak a little bit too to maybe your approach to diversification and risk Management okay Simon so um I view all markets as having their propensity to exhibit these chaotic properties and when I look at um distributions of returns of the market so I all see that these liquid markets over large data sets has what I call

Fat tailed regions they've also got Peaks around their equilibrium which is where the predictive Traders Harvest their opportunities but in those tail regions Which most predictive Traders want to keep away from because it's more chaotic out there in those particular zones I apply a trend following model and I apply a diversified Suite of trend following models and across a diversified array of markets to extract these opportunities that chaos does present from time to time with these enduring Trends what I call the outliers and when I talk about Outliers if you could imagine these trends that

can might last two to three years that most predictive Traders when they are trying to navigate these Trends they're always saying oh the trend's got to come back down it's got to mean revert it's got to change but I'm sitting on the other side saying they're more uncertain than you recognize and I'm going to keep writing this trend so these opportunities as you are aware Are a fairly few and far between so the way I I need to address this is through intense diversification across markets and also not participating in the Market at all times

so my models have to turn off and they've got to turn on where they turn on is where my assumption is that as price gets more and more extreme we find the predictive modelers tend to assume price is going to revert now when these price models get to that extreme level where we have material Trends that have been going for a fair while that's where the predictive modelers are saying I'm going to mean revert here because it's got to revert back to the equilibrium I'm saying No this is the time I participate so I jump

onto these Trends right at the time the predictive modelers are saying it's going to go the other way which is usually after a fairly large price extension so that therefore means that my look backs of my Trend following Models have to be medium to long term I'm looking for significant material price moves and the way I see these markets and their fractal nature going from the small to the large is that as you go to the medium to long term these fractal structures of these enduring Trends get more and more prominent and the noise associated

with those enduring Trends gets less and less so I find higher signals Out in the medium to long term for the specific types of trend I'm trying to Target which are these outlives how how long on average would those trades hold a position for so on on average they might hold eight to one eight months to one year and their maximum hold might be three four years um so I'm definitely dealing in the the medium to long-term space of investing and trading and how many markets do you end up Diversifying these strategies over and is

it more or less the same strategy on on on each marker each contract yeah so um the models I deploy I diversify across my system so I might have 10 different um Trend following systems and what I mean is that each of those systems has golden rules applied to it for Trend following they all cut losses short and let profits run but they do it differently each one of them does it Differently and each one has a different entry signal so this gives me what I call diversification of system and that's I then apply those

Diversified systems as an ensemble of systems into this chaotic environment which I find that Ensemble methods of forecasting are much more realistic than single trajectory predictions in these chaotic environments because with these Ensemble methods where you've got multiple different approaches being Applied to these chaotic environments you find that there is a greater expectation with The Ensemble methods that overall you're on the right path in these chaotic regimes yeah and the markets are extremely diverse as well so because I believe that these what I call these leptocurtic signatures of any liquid Market offers these Tails my

models because they use small bits um a stop and a trailing stop simple forms of exit which mitigate risk I can Afford to trade what others might think are fairly highly correlated markets but I'm doing that because the interaction between my systems and those markets means that if I'm trading crude oil or Brent oil which are regarded as highly correlated markets I can get outliers in one and not in the other simply because of the interaction of my Ensemble of trend following systems has with those markets and in our world of chaos where you might

have heard of the Butterfly effect even though we get markets set over the course of time are highly correlated very small variations of those markets are sufficient to create outliers in my trade outcomes and this is because the the small variances in my very non-linear world can amplify with the trade outcomes and this is a different proposition to a predictive Trader so when things become very predictable they typically are around a condition of Equilibrium When Things become very unpredictable they are typically diverging away from equilibrium it's you know in physics we see this as phase

changes between liquid ice to water to gas we see this transition events occurring between these different ordered phases and those transition events are my Tails this highly disruptive state that exists as a system is transitioning out of one ordered State one predictable State into Another ordered State and the necessary transition causes disruption to the prior State and then a new state is resurrected once that transition is over and I see this in the financial markets and that's why in 2008 we see this massive decline in the equities Market of say 50 percent and at the

end of that phase we can see a reversion to a more predictable state it it's the transitions over and and the way I see this is if you could imagine I talked About the 90 10 mix the predictive 90 and the 10 unpredictive what happens when you start getting into these unpredictable regimes and these predictive modelers find that their models no longer work they're forced to become price followers like us because they have to exit their positions and when they do that they start amplifying our signal and say the 10 slowly builds the 20 30

40 50 when we get 50 and then we get further above that's where the Trend ends because we're looking at the asymmetry that exists between the 90 and the 10 and how wealth flows to us when the predictive modelers get it wrong so can I just Loop asymmetry yes just a real quick loop back to your um background and and working with an re and and talking with other managers and and your experience there did that um like what did you learn from the way the the traditional fund managers were Trading because that sort of

relates to what you're talking about now I suppose um what would take home to me was was the the people that I really strongly investigated at that stage of my life were what I call predictive modelers say value investors fundamental investors in other words they do this incredible detailed analysis and from that analysis they determine what they call this intrinsic price level that their analysis concludes is that price level They then look at where current price is in relation to that intrinsic price to determine whether it is undervalued or overvalued so that intrinsic value is

an intrinsic um predictive point in the future that things will converge to so I view their their models as convergent and predictable I've got to sit on the other side of that and say that yes that is the case a lot of the time most of time but because I'm concerned with um markets displaying this uncertain ability which is fundamentally unknowable um I will therefore have models that are taking the counterposition to those predictive modelers but be because I know that predictability can rain for very long periods of time I cannot afford to not mitigate

my risk in these predictable environments so I'm continually cutting losses which has given me lots and lots of small whip Saws but I'm never letting my losses get to be non-linear Adverse Events they're all small random perturbations lots of small ones but when I get it right with this Mass transfer of the predictive model is getting it wrong and suddenly these these Trends explode and and I'm writing them far longer than what a predictive person will ever assume or what what gaussian models would ever express you so is that the primary risk management Defense in

all of its Simplicity just that diversification and the um and the uh the The Cutting losses and letting winners run overall across all those instruments yes incredibly simple isn't it and this is this is this is what gives me such a big buzz in that um I believe um that the power is had with the simpler approach in these chaotic environments of course that makes sense Because you think if I have advanced super complex models they're trying to predict things with too much accuracy there's not enough freedom of of being wrong but with simpler models

which are far less optimized for particular future targets that gives freedom to move and therefore with these chaotic features that I'm trying to draw as much of an edge from those chaotic features as I can the simpler model the better and which means that I need to mitigate Adverse risk all the time but let my profits run so what that does to my trade distribution is create what we call a a strong positively skewed signature and where I see the predictive model is facing a problem is the negative skew that's inherently embedded somewhere in their

premise so I'm sort of on the counter side to the negative skew with the Positive skew if that makes sense so how simple is too simple rich like The The reality is that um uh making money in the markets is difficult it requires a lot of time and energy not everyone can do it so what is the um what's the thing that makes the difference between a simple thing that anyone can either do or buy off the shelf and actually being consistently profitable because the Simplicity is naively simple it's actually incredibly complex how you Come

to that simple solution and uh the way I view that is that you know we we have simple equations to describe very complex phenomena general relativity that's a that blows my mind that's actually a simple equation and then you know Einstein came to this conclusion that when they talk about parsimony and when they talk about Occam's razor what what the physicist is saying is that um the whilst being simple what that means is Their explanatory power is much greater than the more complex systems which are much more uh specific so when you're trying to explore

new domains and you're taking your your models you've developed for very familiar domains like the domains around this Earth and Newton um that the the scientist Newton developed these very simple models to explain gravity and and masses and movement in this simple domain he wasn't he wasn't aware of this large domain out There which challenged some of his simple assumptions and his models Einstein came along and said that's a very good localized solution Newton but it's one solution and there are more comprehensive Solutions when you have to start looking at things such as black holes

and complex curvatures and things that Newton wasn't even concerned within his small domain and so Einstein revised Newton's propositions to create something that totally blew everyone's Mind because it wasn't just a revision it was a total reforcing of this idea of of what Newton had about time being static space being static Newton blew that apart with his dynamical models showing that space and time were relativistic content Concepts and how and all of these things it led to this incredible expansion of understanding but his his model was very simple and his fund this formula was very

simple that's the way I view these markets so Whilst I say uh Trend following models are very simple it's naively simple because there's a hell of a lot of work I've had to do in my research to get to where I am now with these simple models under diversification all right well um let's move forward from there then and um and tell me a bit about how you do get to those models in the sense of your strategy creation process um you know say generating an idea Testing it do you have a process for that

testing and then how do you make sure that you're not fooling yourself and curve fitting to the data when you do test those ideas great Point Simon Okay so the fundamental principle or assumption that I use for my models is the assumption that these outliers these these indeterministic chaotic regions are prevalent in all liquid markets that's a big assumption and I've tested that Across all the liquid markets I come into contact with and I test that by looking at very long-term data sets and those data sets all conclude that they have this leptocurtic signature so

as soon as I see that they've got fat tails in that I know they're dealing with what I call non-linear mechanics and as soon as I know that they're dealing with non-linear mechanics I can go to my scientific fallback that non-linear mechanics leads to chaotic Structure that is fundamentally unknowable indeterministic so knowing that gives me so much certainty and I test that um also in the back test environment so when I'm testing my what I call these 10 Trend following systems these ones I've told you about they're universally applicable across all liquid markets what that

means is I'm not looking for characteristic patterns that each individual Market might have I'm looking for a universal principle that is found in all those markets but it's not a pattern it's a universal principle over very long-term data sets so I'm not targeting patterns which means I can choose to adopt very simple models that don't curve fit don't over optimize because over optimizing or curve fitting is a condition that is found when you rely on the the signals or or the the inferences contained in your back test to project in the future And when I

look at a particular Market it does have characteristic signatures and mean reverters will deploy those characteristic signatures technical analysts will deploy those characteristic patterns AI will deploy those characteristic patterns into the future it's this learning exercise what I'm saying is no that's a world that I don't wish to compete in that's a predictive world I'm looking at these Universal properties these Tails which Are unpredictable so how do I trade an unpredictable price pattern and I've got to take a bet either a long or a short bet but I've got to in that decision cut the

other side off short so if I take the short bet something that is fundamentally you know indeterministic if I take the short bet I make sure that I have my my trailing stop and my Stop close to the adverse proposition to that directional bet I've taken vice versa for the long bet and And by that very simple model giving lots of freedom to move I then just ride the uncertainty which most of the time gets me losses whip saws but some of the time it gets me 180r or R moves and R is a term

we use when we we strike an entry price and we put a stop in place uh the distance between the entry and your stop is one hour so I've had instances in my back test where I've got models that some of My Trends last 100 are over two to three years the these indeterministic things that people this is where these outliers can deliver monstrous profits and so when I look at my trade distribution of returns I see that 90 of those returns are basically dealing with the noise of prediction dealing with my models being wrong

because those models have been that have been applied in environments that are fairly predictable And I've been wrong in my assumption ninety percent of them but because I've constrained my risk for every one of those bets that I've placed in those 90 I've limited my adverse risk but because I've let my profits run in the 10 of chances that exist in my trade distribution they've been King hits so under this massive diversification these King hits these non-linear King hits because I've got say a hundred R compared to a normal loss which is 1R You can

see the the non-linear relationship a hundred times that small loss that's where Ed sakota says one big win pays for them all I might have five to ten percent of my distribution being the big wins they pay for um the more the 90 which can be wins more wins and small losses that's how Randomness goes and that's how noise goes but they can be so big that's what creates my wealth trajectory over the long term um presumably then you you're trailing a Stop along that so you would still need to would you not test out

a little bit um you know just how far behind you want to trail that stop and how much you're prepared to give back so is that a sort of a traditional um uh just machine learning approach to to finding averages there or how do you keep that simple it's it's not simple and the way I've Test that is through massive data set testing massive dark lab so I'm not only testing my models on one market I'm testing them across my entire portfolio that might be 70 markets 100 marks whatever it is so because I'm looking

for this Universal principle I'm looking at massive um sample size massive sample size with my trade distribution and if you can Imagine I described what I call it the the typical leptocurtic curve at the markets we've got these fat tails and this sort of sits outside what we call a normal distribution so with my data sampling and where I set my Trails or where I set my stops I'm trying to find the boundary between the gaussian distribution and the non-linear distribution so I'm trying to find a raise a boundary where things become non-linear from linear

so in the under The Bell distribution the bell curve of a gaussian distribution that's a linear distribution um you can get that through flicking a coin an unbiased coin when you flick the coins and you plot your consecutive wins and your consecutive losses that plot will follow a normal distribution a random distribution but leptocurtic distributions means that it's not a game of chance peak of a leptocurtic distribution there is an area which is exploited by the Predictive Trader around an equilibrium around an assumption of a future State around this equilibrium at the Tails this is

another area where an edge can be exploited which is the area I'm focused on with my particularly focused concentrated technique at those tails and I'm looking for the boundary of normality and the boundary of chaos does that make sense yeah and so where I place my stops and trailing stops with Massive sample size says hey this is a this is the boundary of where linearity meets non-linearity this is where I'm going to stick them boom boom down they go and they will change over the course of of the trend because their volatility adjusted as these

stops and trailing stops as they go over the course of a trend that I'm using what I call a chandelier exit technique which means that your Trail moves behind price but is allowed to Breathe under high volatility and is allowed to contract under low volatility but a Trails price all away so that gives me my my statement that I will let profits run provided it sits above that trailing exit condition and that trailing exit condition will eventually decide when that system decides that that trend has ended and I've got 10 different systems interpreting Trends differently

right but of those 10 different systems They those systems more or less get deployed as they are over any contract or do you tweak things per per contract and just talk to talk to us a little bit about um because obviously I trade stocks but you're really trading either futures or cfds and just mention something about trading cfds in Australia as well sure so um because I'm on the hunt for maximum diversification with my models and the Reason for that is that I find that my um the number of outliers I get in my distribution

increases with more diversification this is a different to the predictive modeler who assumes that there is a certain level of diversification which is optimal and beyond that diversification gets more and more marginal and the reason they're doing that that is that they're not they're they're looking at things in a what I call a homogeneous way Assumptions in a homogeneous way they're assuming that all markets display similar behavior in general efficient market hypothesis sharp all of these things are based on these simplified assumptions of how markets behave but it's not the reality in my non-linear world

I know that small perturbations can make massive differences so in my hunt for outliers specifically because I'm targeting outliers as opposed to a predictable condition I know that I can Get maximum diversification to achieve that because I've got finite capital and I'm not a Jerry Parker at this world he's he's got you know millions and millions as fine art Capital I've got far less Capital the way I do it in-house with my models this is outside of my job with East Coast Capital Management which I'll explain shortly but in my my operations as a retail

Trader effectively with finite Capital I'm Using cfds because they give me maximum diversification potential as opposed to Safe Futures which means that unfortunately Futures don't give me the ability to trade micro Lots and because cfds give me the ability to trade very small bets that means like a massively Diversified both in markets and systems just remind me you pay a holding cost of um um a cost of carry more or less for trading cfds do you not and does that Impact especially a long-term trading strategy depends depends on the markets you trade and your interest

rates at that time but you can get positive carry with going short you can have negative carry going along the interest holding costs can be beneficial to you what we call positive swap or they can be negative two called negative swap that's important in our cfd world and you've got to take the inherent Differences of the markets we trade in cfd land versus Futures because some of those impacts of transaction costs are different and you've got to assess that in the models you develop to trade these markets because there are inherent differences typically in cfd

land because um it's it's it's done by brokers who act as the counterparty to your trades in the cfd world they're market makers for the cfd world you will Find that there are the Brokers have specific conditions that need to be met for the Traders which necessarily mean that you might need to trade these models slightly differently to how you're trading other markets and typically the impost of trading costs in the cfd world is higher than Futures you're trading higher on average holding costs higher on average spreads higher on average commissions this is all for

the convenience of trading cfds and so You've got to balance up the additional transaction costs of the product you're trading like cfds against the diversification benefits and the convenience of trading those products which are cfd is when I do that evaluation I say yes I recognize I'm paying higher costs but the benefits of diversification are so extreme this is the the product for me to apply while I've got small finite Capital when I become a big boy I view what I'm doing In my retail world it's a bit like a novice with trainer Wheels in

an environment and when I become a big boy with money at AUM under under me then I'll be able to trade Futures which has lower transaction costs and in the scheme of things that's where you want to be in that Futures environment for our particular type of models now are you using um Ai and you have spoken to me a little Bit about um some projects with chat GPT that I thought were very fun can you um can you Enlighten us a little bit on your experiences with chat GPT sure I'm I'm embracing chat gbt

as another fantastic tool um so I loved it when spreadsheets first came out I loved it when algorithms came out I love it when programming came out and I'm loving it with AI comes out it has this ability to really interrogate Big data and come out with some pretty great ideas yeah using that approach so the approach for instance I used was that um in deciding a possible Universe to trade and this is an example where I said to myself right let's assume I wanted to apply my models to stocks now one of the the

key assumptions of my models is that I'm trying to find uncorrelated markets to trade that gives me a diversified level a little extra Edge because these uncorrelated markets Produce risk offsets which reduce the volatility that I naturally get chasing my outliers so I do like uncorrelated markets and I say the check gbt hey chat GPT can you please scan the S P 500 and select 20 markets or 30 markets that are uncorrelated using logic alone not using correlation properties using a logic alone Chef GPT comes back saying sure can and it does its analysis comes

back with These 20 to 30 stocks and I said please list these and the reasons for your your choosing under your logical framework and it comes out with a stock in healthcare a stock in AI a stock in here a stock in there and it says it gives the reasons of the functional reasons in logic space why they are uncorrelated and we say well that's great but the good thing about you've always got to validate what comes out from it because it might sound great But you've got to validate it so then I said I've

got chat gbt to validate itself I said right you've selected these ones now use data that you've got available to yourself and says I've only got data up to 2021. so we'll use that data and um use it to evaluate on a correlation basis your stocks you selected during using logic to see if they have these uncorrelated properties they said cell can comes back and says correct my logic is sound it Demonstrates that uh not only through logic but also through the correlation properties of the data that I've assessed over 20 years 30 years or

whatever they are uncorrelated and I go great I said fantastic and then you Chuck that into your back test environment and then you find you've got something that delivers returns well in excess of the index with much lower volatility there you go in 30 years there you go so it's just an idea It's just an example of how you use AI to to expand your capabilities you want to leverage off your computers leverage off these tools because we've got a very creative brain that does things incredibly well but it doesn't do some things very well

lots of crunching lots of processing Big Data brain's not too good because the brain likes to heuristically try and resolve things and all these biases come in so this is Where Chet GPT spray so in in building these Sim simple if you like models coming back into that um they they're obviously there because they're enduring you expect them to last a long time is there is there a um a framework in your world for assessing whether a particular model is just no longer valid or or needs to be retired is there that kind of process

or really Things Are There to be used because they are long-term and enduring they're expected to continue to work no matter what yeah so it's a bit different in our world to say the predictive world with predictive models um what they do is they typically use in-sample data and then they see how it performs in outer sample data and if it falls off the cliff as soon as they go into the out of data sample they say that's a bad model strike it off the List we won't go with that anymore we don't do that

because all of our models have been rigorously tested with using historical data and the two things that they say is how have they how are they performed when outliers have been present in that historical data if we find there is a positive correlation to those outliers that's a tick how have they performed to other risk events so things that long predictive Stretches all of these things that's where we're looking at things like adverse drawdowns the adverse side of the risk it comes back and say yes it is optimally operated within those adverse risk environments that's

a big tick to our models because we're going to be putting these models into a future environment that we hope is uncertain so if there are future outliers there's a good chance our models will catch them And if there aren't there's a good chance that our models won't fall off the cliff but we've got no preconceived idea or expectation about how they're going to perform because we are dependent on these outliers that are by definition unpredictable in nature so the way I do this is I don't for instance have a threshold to say if I

achieve a drawdown of 40 on a particular model I'm going to eliminate it from my basket of models Because I know it's past the test this is just an unfortunate instance of a particular predictable regime that's lasted longer because I'm confident that they will capture those outliers if it occurs in that market but because I've Diversified the impact of that adverse risk over the long term small losses building up is small for the portfolio because it's so Diversified so the way I do this is every year I add new data to my incredibly long data

set Testing across multi-markets and many years of History so each year I'm generating new algorithms to trade the following year each year as more data comes in that's giving me more possible historical outliers to test more possible historical conditions that haven't been seen previously in the back test to test against so it's a type of robustness testing where I'm using an extended data set to regenerate models so at the end of each year I use a whole New fresh batch of models whether or not those models have performed well or not in a prior year

but what I do do is I allow those models to run to their completion but I supplement them with new models each year it's a bit like evolution in action if you can imagine what I'm trying to achieve there that's an interesting approach I like that yeah all right so um I guess just just going fine just Starting to wrap up but just to go a little bit deeper into the strategies while we're here um uh your um your building strategies that uh that that that that are enduring and then you'll review them with the

new data that that comes in um that whole so I guess going back to the The so and at the end of the year when you're updating those models is there a new idea generation process like if you're if you're reading and learning through the year as well like how much can things change and and how do you continue to generate I guess those ideas for those updates so as as my learning improves my um I'm always sort of looking at new models uh so I've talked about the 10 Trend following models I'm using you

know I Will um look into new um models if I find that I think that there might be opportunities that are left unturned with my existing Ensemble but I must admit um where I'm at now has been something that's taken a decade to achieve and I'm incredibly confident with my process now what I'm finding with the research I'm doing more and more as new research comes out is it it's it's giving a tick Box to my process not requiring me to change it um you know I will always Embrace ideas if I find that they

are better ideas better models I always Embrace that but it's getting to a point now where through Occam's racer through persimony I'm getting down to sufficiently complex models in other words not too light not too complex you know what I mean that that balance I'm pretty happy with that you don't need to tweak it much so You've got a very strict process for that um the annual update is that in itself quite codified and strict is that almost an automated process or what does that look like yeah so the entire process of portfolio construction from

the ground up we've tried to make systematic so all rules-based algorithms that create the process so most people are familiar with algorithms that they can get an Algorithm and then apply it to trade something we use algorithms in the construction process to get to those end stage algorithms if you know what I mean so there's a workflow process we use where the first thing is we say right Fred the coder that I use will will I'll say to him Fred go out and develop 10 uncorrelated Trend following models and he'll go out and I'll say

also let's keep it simple maybe have five or six parameters at most let's make sure that The golden rules of trend following are embedded in it and you come back with your design Solutions Fred comes back with 10 say design isolations then what we do is we say right now let's put it into the systematic environment testing environment where we'll say these Trend following models because of our particular assumptions must be capable of trading any liquid Market we present it with so then what we do is we say let's get a data set of Say

200 markets to evaluate our models with we will then say in with those models which are Trend following models we want you to determine across all 100 markets every single one of them what are the parameters we should use for or the variables to use for those parameters for all of those 10 models so if you can imagine that is a massive multi-market test where it is testing not only on an individual Market but the entire 100 Market data and it's coming up to these these variables that is finding what we call in the edge

of chaos and and the edge of normality where where that edge lies that process comes up with those variables so then we say right there the trend following models now let's what we call visually map those models to Market and this is where we test for over optimized models so fortunately in our Trend following World outliers a Visually very easy to see they stand out like proverbials when you are looking at Market data and you see where an outlier is it's obvious so we then what we call map to Market our models does the model

perform well in The Outlaw region and how does it perform outside of that outlier region if it performs well in the outlier region that's a tick if it performs well outside of the outlier region we give it a cross because if we have Trend Following models that work when markets are constrained and not offering Trends there's got to be something wrong with a model so what we're looking is the correlation sorry just to just repeat that so I understand um the the model should deliver in the way that it's expected in the market the kind

of EX Market that it's expected to perform well in and if it's not a favorable Market it shouldn't perform well otherwise it's right and if you Find something that doesn't tell you that suspect overfitting this is where your models might be fit to random noise as opposed to the signals they're trying to extract from that market yeah so also so I've got very simple designs we test across multi-markets we've got this map to Market process which is all all three of those are aimed at reducing the ability of our models to be over fit to

a specific characteristic of a market or noise to a Market all of these are our way of robustness testing our models and then at the end of this process therefore we might have 2 000 solutions that we've only got finite Capital to deal with how do we collectively compile those Solutions into an optimal portfolio Trader and the way we do that is through a process of not not through correlation studies looking at the individual correlations statistics of each return stream we're Looking at the entire return stream of each of those systems and we are iteratively

bundling them up into groups of say 40 so let's say we've got finite Capital to trade 40 markets which of those return streams out of the 2000 are we going to trade we use an iterative process where we are bundling together for 40 possibilities out of that 2000 selection to produce the optimal part dependent risk metric ma ratio so that you can imagine this is a very crunching Computer intensive processes that's working over billions and billions of iterations to come up with that and once we come up with that we know there's numerous tick boxes

in that entire process that is tested for robustness tested for over optimization tested of fitting we're very happy to take that bundle of 40 strategies as our next year's Edition for our Trend following portfolio and then next year more data comes in we do The same process again Etc so have I understood correctly that one thing that could change year to year is the actual markets traded yes yes absolutely and um if you're comfortable now ideally Jerry Jerry would be trading all markets but remember in our finite Capital world of a retail Trader we can't

do that we must make a selection unfortunately and so this is how we select if you're comfortable just discussing maybe or Giving us some insights into the software and the data that you would use to to go through this kind of process so um there's off-the-shelf software such as um strategy quantics adapt trade there's a few of these uh what we call data mining software but you've got to be careful with data mining software and how you use it because it can develop very over optimized Solutions so but within strategy Quant X there are Particular

processes a trend follower can use using what we call a custom project functionality of sqx to rigorously assess and then compile your portfolio through an iterative workflow process like I've described but Fred and I have decided to do it ourselves outside of that environment so we use just a simple platform that is tradable such as metatrader mg4 and then Fred develops coding outside of the mt4 environment that then uses the engine of Mt4 to undertake the tests but what we can do is we can set the whole process workflow process up where we click a

button and then three days later come back to a compiled portfolio yeah wow that's fantastic and that gives you a lot more control yeah that way all right well we should start wrapping up we've just broken the hour Mark um tell me a little bit about Um in in tell me a little bit about educating the average investor about quantitative Trading and um and whether they should be more interested in allocating to this perhaps they don't even know about it and and as part of that just tell us a little bit about eccm yeah so

systematic trading you and I both stand in the same court here we we advocate for systematic trading um In my world of trend following where you're dealing with thousands and thousands of return streams you cannot do that in a discretionary manner you must have a systematic rules-based process to apply that because we cannot operate at the speed computers can do to run our models with Fidelity run our models according to our back test systematic is also essential because it's a way in science it's it's akin to having a hypothesis or a model and then Empirically

testing that hypothesis using data using what's available historical data as a way to empirically validate that model I've talked about Newton before and then Einstein so Newton had his data generated from the Earthly planet and his models were developed now very successful models used for hundreds of years based on the data that was available to this planet unfortunately the models were not precise enough to account for things Such as we take for granted in GPS satellites now to give us the specificity of of location Target things that can't account for gravitational anomalies such as black

holes massive scale structures things that can't account for the smallest microscopic structures you find in quantum physics so in in that broader domain Einstein started using more data from you know Venus from black holes from solar system so it's it's science has continuously Empirically validating its models against the data that's what we do in our world recognizing that our domain might be small to start with but as you increase your domain more and more data is required to validate your hypothesis that's a systematic side now what was the other question you asked um tell us

about eccm I know you do so okay so I am what you call a strategy Ambassador for East Coast Capital Management they are a sydney-based fund manager that Specializes in in my coursework Diversified systematic Trend following so I work closely with the managing director Adam haverlive and um it's uh We've currently just generated an information memorandum for the fund it's had a uh a track record since um January the 1st uh 2020 up to current day but now it's being umbrella with an information memorandum to expand on and we're looking to attract investors to to

that currently So very excited about that all the rules I've been talking about that I do personally in some way shape or form has been umbrella into Adam's work and we're very much aligned so uh yeah yeah that's excited and and they've done really well so that is exciting um all right Rich well um look just to finish off going from the um The institutional side to the retail side if you're just to give some advice to someone right at the other end perhaps just starting off or early in their trading systematic trading Journey any

advice for them I would say my advice to anyone is continually question never adopt never adopt something learned from elsewhere and assume it's valid you must test it yourselves to not only Develop the confidence but also make sure that it marries with your brain the way you think the world Works um you there's a learning that goes on with trading where you're continually building your your brain you're building your strategies they work in tandem together so it's got to be a Personal Achievement of self-reliance not dependence where you get dependency there's going to be failure

so read the books read the hard books not The easy books read the hard books read the challenging books look at uh look at what's currently being produced by economics as questions not as facts things you want to test for yourself validate for yourself they're the things I'll be be basically saying as advice and get a mentor get mentors that are you want to stand on the shoulders of giants so get those giants understand what they've gone through because it's not an easy game it's Incredibly hard and you need those mentors to give you guidance

wonderful all right well is there anything else I've missed Rich that you want to touch on absolutely nothing Simon complete that's brilliant and I really appreciate that that was a great um Journey Through the scientific process and approach to trading I enjoyed it a lot um how would I suppose if people want to get in contact with you or myself it's Really the the same thing they can just simply start at our website for the podcast the algorithmicadvantage.com um all right Rich let's leave it at that thanks very much and we'll see you on the

next one many thanks Simon and see you all bye-bye we should remind you that the conversations on this show are informal and for entertainment purposes only certainly any general advice you may hear is obviously not specific to your Needs goals or objectives so nothing discussed on the show should be considered as investment advice if you want that you'll need to actually do your own research and speak with your financial advisor remember trading can be extremely risky in past performance it's not necessarily indicative of future returns if you enjoyed the show Please Subscribe or leave us

a review and if you have any questions or feedback we'd love to hear from you bye For now foreign foreign [Music]