I received a nice email from a client who had just listened to a podcast I co-hosted with Mark Longo at the Options Insider last year and my client loved it. I went back and gave it a listen, Michael Mescher from Gammon Capital does a great job as guest.
- What is Gammon Capital?
- You've talked a lot about the dangers of selling volatility and how short vol managers aren't worth their fees. What are your thoughts on the maelstrom of the past week? How have your funds fared over the past week?
- Machine learning is a hot buzzword these days, but what are some of the practical applications that you're seeing in the financial markets right now? Is it ready for prime time or still in the nascent phase?
- Unsupervised learning/ fundamental strategies?
- Supervised learning / quantitative strategies?
- Classification vs. Regression problems
- Hiring machine learning specialists vs. Products specialists
- 90% of the world's data was created in the past two years. Is that an opportunity or a terrifying statistic?
- Testing models on out-of-sample data vs. carefully constructed back tests?
- Should managers be concerned that the robots are coming to take their jobs and their clients?
Michael Mescher heads the team at Gammon Capital which he founded in 2015. Mr. Mescher is the Chief Investment Officer of Gammon Capital’s program offerings. Over the past 15 years, he has established himself as a successful trader with extensive experience in listed derivatives.
Prior to founding Gammon Capital, he was a partner at Ronin Capital in charge of managing structured volatility strategies. His experience also includes 5 years on the sell side at Lehman/Barclays where he was the Head of Special Situations Volatility Trading.
Mr. Mescher began his derivatives trading career on the floor of the CBOE. It was there during his free time, that he began applying statistical analysis to backgammon money games with great success. It is these methods and processes that inspire the approach at Gammon Capital.
Here's the interview:
- You're listening to The Options Insider Radio Network.
For more quality Options programs, visit www.theoptionsinsider.com or search for Options Insider Radio Network in your podcast provider of choice.
Listeners can also access all of our programming through our mobile app available in iTunes and on Google Play.
Don't forget to follow along with your favorite programs and submit your own questions for the host at twitter.com/options, facebook.com/theoptionsinsider or via questions at theoptionsinsider.com.
- Welcome to Trading Tech Talk, the program where we break down all the amazing technology that takes your trades from the click of a mouse to the clearing house and back again.
Each episode, we'll break down the latest developments from the world of Trading Tech, bring you exclusive conversations with industry thought leaders, answer your pressing technology questions and much more.
If it involves the tech that powers your trading, then you'll find it on Trading Tech Talk.
Trading Tech Talk is brought to you buy Trade Alert LLC.
Trade Alert tools, sets the standard for accurate timely Order Overflow Analysis for sales and trading professionals worldwide.
And now, it's time to break down the world of financial technology.
It's time for Trading Tech Talk.
- All right everybody.
That music means it's time, once again, for Trading Tech Talk, the program here, where we break down the ever complex sometimes ever confusing world of trading and technology and explain how you guys use it (laughs) executing your trades and bringing those trades all the way from the click of that mouse, to the clearing house and back again.
My name is Mark Longo from theoptionsinsider.com as well as, of course, from the ever expanding ever exciting, Options Insider Radio Network.
If you like this program, make sure you go and wherever you listen to this, could be iTunes, could be our website, could be via our mobile app or any of the above, make sure, if you haven't done so already, make sure you upgrade to the full network feed so you get not just this program but of course the other dozen or so programs you could use on the network.
Which I'm sure if you like this one, you're probably gonna get a kick out of as well.
And of course I hope you listen.
And whether it's with iTunes, the site, however you got the show, make sure you send in those questions those comments.
We do like to hear from you guys.
You guys help set the tone, for what we're gonna talk about here on the old program.
And joining me on Trading Tech Talk for our first episode here kicking off here for 2018 is an old friend to the network.
He is Mr. Matt Amberson, the founder over there at ORATS.
Matt, welcome back to Trading Tech Talk, sir.
- Thank you, it's great to be here.
I'm excited about our guest today, Mark.
- Yeah, we have a good guest so let's get right to it with our CTO interview.
- It's time for exclusive conversations with the people who are reshaping the world of trading technology.
It's time for (beeping) a CTO interview.
- All right everybody, welcome to the CTO interview, the portion of the program where, as the name implies, we interview guests from throughout the world of trading and technology and pick their brains for the benefit of you out there, the listener.
As our next guest pointed out, as many of our guests have pointed out on this segment, they're not technically CTO's but we won't hold that against them because we're really here to talk about the nexus of what's going on in the trading and technology landscapes and how it impacts the world we all live in, of course.
And our guest today is Michael Mescher.
He is the founder over there at Gammon Capital.
Michael, welcome to the Trading Tech Talk program.
- Thanks guys, appreciate having me on.
- Well Michael, as we are wanting to do with all of our first timers here on the old network, why don't you go ahead and give us an overview of your background in the trading technology landscape.
And then, what the heck it is that you guys do over there at Gammon Capital.
- Sure, so just a quick background.
I've been in the industry about 20 years now.
Originally started on the floor of a Cboe with one of the leading market makers down there.
Moved from Chicago over to New York to join Lehman Brothers in January of 2008, which was exceptional timing as a--
- Auspicious timing there.
Well done, sir.
- I'll tell ya some stories after the podcast.
Let's see, as you all know, Barclays took over Lehman.
When I left there, I was running special situations volatility trading.
Jumped over to the buy side, and then, in July of 2015, I started Gammon Capital.
So, we're mostly doing quantitative trading on the options side.
And probably about six months after starting the firm, we had acquired a fintech, that had specialized in derivatives analytics which I think we'll probably dig into pretty well on this podcast today.
- I didn't realize you were a former Cboe guy.
This show lousy with a lot of former Cboe guys.
The network indeed lousy with a lot of former Cboe guys.
Which pits were you in down there?
- I was in the S&P, let's see the SPY's.
I was working at CTC so they got the SPY pit when that got listed.
Also the Nasdaq products.
NDX is where I started, so.
All the big industries.
- All over the place.
We'll I know you've been active.
And we're gonna get to all the machine learning stuff in a couple of seconds,
'cause I know that's what you guys are doing.
But I would be remiss if I didn't kick off the interview,
I know you've also talked a lot in some of your appearances and some of the things you guys do over there, about the dangers of everyone flying into the short volatility trade and how it kind of works until it doesn't.
We saw the latter side of that coin happen and unfold before our eyes last week when we had a very tumultuous,
I think is a euphemistic way to put it, a very topsy-turvy week in the markets.
I'm curious, what are your thoughts on the week we just witnessed and all the legions of short selling vol funds that perhaps are no longer around including some products?
Like XIV, that are about to head to the graveyard as well?
- Yeah, for sure.
I think probably every time I've ever lost money in my career,
I've chalked it up and called it a tuition payment.
And I think anybody that's ever traded options, especially from the short side because that is typically the first trade that anybody learns as a professional options trader 'cause it's the easiest, you learn that brutal lesson of what happens when the market eventually blows up.
So, your really seeing the landscape change with the invention of these ETFs.
And basically, what happened is the market created products that allowed portfolio insurance to be crowdsourced.
I think that's probably a great idea.
I think that's the natural evolution of markets.
I'm pretty sure Credit Suisse will probably catch some flack for this product.
But what they did was actually fantastic.
I think what actually the problem is, is that people didn't understand what they were getting into.
Selling insurance, there can be money to be made on that if you risk manage it properly.
But so many of these people just got caught up in the easy trade and they had to pay that tuition payment like everybody does eventually.
- Yeah, it does seem like a lot of people got caught up in that whirlwind.
We've certainly seen a lot of the feedback of just to our network over the past week.
A lot more people trading XIV than I initially assumed out there.
Particularly in the broad basic retail parts of the space.
Which is probably where you maybe don't wanna see a lot of people diving into a product that before they really understand all the nuances.
We have, of course, been warning about that for ages.
Really make sure you understand what you're getting into, read the prospectus know that there are these termination events.
These, I think they call them Acceleration Events in their prospectus, it sounds much nicer.
But they do linger over that product, kind of like a sort of (murmurs) dangling over that products existence.
You should be aware of that stuff before you go into it.
Clearly, a lot of people were not.
And we saw the fallout from that as a result.
Really quick, Matt, maybe before we get into all the madness of the fun coming up with machine learning.
Really quickly, and your thoughts on the (murmurs) of last week?
Because it is just so front and center for everyone these days?
- Yeah I think Michael spelled it out pretty well.
But, we did a study of what actually happened to the markets.
And I think one of the things that you and I have talked about is the dwindling team of market makers actually out there.
Maybe five left, and then the
And when the market goes crazy like it did last week, wow the market's doubled in bid-ask width and the size on the bid offer went way down.
And it cost option traders, what I calculate, almost a billion dollars over what they usually pay in the slippage.
And so, it was a crazy week.
Not just caused by this
VIX related and XIV related debacle, but man, it sure cost a lot of people a lot of money, Mark.
- Yeah, I'm looking at some.
You have some great visuals accompanying the analysis you guys did.
You guys check it out over there at orats.com
O-R-A-T-S .com, the blog there for all the visuals.
I'll try to repost it on The Options Insider for you guys to get at it too 'cause it's a piece
I think a lot of people should see, just because.
I think when you see the visuals, the contrast is pretty stark.
We could talk about this for a long time but I wanna dive into the core of what we're here to talk about today.
Michael, as you probably are well aware,
I think to call it a buzzword is perhaps not doing it enough justice.
It's everyone, and their mother, these days outside of the weeks when XIV implodes, seems to be talking a lot about machine learning and AI and the use cases, the implications for the financial markets.
We're seeing a lot of smoke blowing around these topics out there.
Not really seeing a whole lot fire yet.
So maybe let's start there.
I'm curious, you spend a lot of time over there at Gammon researching and looking into these topics.
Are you seeing a lot of fire right now or is it still a lot of smoke in terms of machine learning?
Is there still a lot of people talking about it but not a lot of rubber hitting the road yet?
- I think there is a lot of both going on right now.
Obviously, machine learning is a buzzword that a lot of guys are really using very heavily their dex.
And on some level, a lot of them are using machine learning.
When it starts coming to regression techniques and analysis, just with the power of Excel you're able to deploy some machine learning right there.
So I think a lot of people are probably talking about that.
But when you talk to an investor,
I think the investor, when he hears machine learning, he's thinking more in terms of deep learning.
He thinks about watts and he thinks about AlphaGo and those type of algorithms.
So, when you start talking to machine learning guys,
I think it's really important to distinguish between the types of machine learning that they're doing and how they're applying it.
Because machine learning is really a whole host of different algorithms that you can use to accomplish basically coming up with answers to questions that you have.
So there's definitely some fire there.
There's definitely some smoke there.
I think a lot of it comes down to just being able to ask the right questions and being able to vet the guys who are holding themselves (murmurs) as machine learning managers.
Because for a lot of the people that I've seen in this space, it's almost like the blind leading the blind.
So unless you really know how to vet who you're talking to, it's possible that you might have smoke or fire, it's not quite clear.
- Maybe we should start there then.
'Cause obviously it's a pretty broad umbrella, that umbrella of machine learning.
Let's drill down a little bit.
When you're talking specifically about machine learning and perhaps, the subset of it that you think is most applicable to the markets here, which subsection of machine learning are you really focusing?
- So, we're focusing on quite a lot of different machine learning algorithms.
It's really tough to say one specific one.
And it's funny that you asked the question because I've seen this with some incredibly sophistic allocators that we've spoken with.
They expect people to be specializing in one specific type of machine learning algorithm.
Maybe it's deep learning, maybe it's some sort of unsupervised learning, maybe it's supervised learning, maybe there's a specific library.
But if you're gonna do this right, the real way to do it, is to take a look at the question you're trying to answer and then go out and pull in the tools that'll help you really answer that question.
And for investors, the type of investing you're doing is very specific to investors.
So, for some examples,
I think everybody's heard the example of taking a look at satellite images and using that as part of your investment thesis.
Okay, those are classification problems.
Those might work fine for what they're trying to do.
On the other side, once you start to venture more over into the options side, you start getting a lot more quantitative.
And maybe the classification type of problems aren't necessarily what you need to be looking at.
Maybe you need to be looking at more of the regression type of problems.
- It's funny you mentioned that.
I went to a talk not that long ago, with I think it was a Google executive talking about machine learning in the markets.
That's pretty much what they went on the whole time about, was classifying satellite imagery essentially.
So it seems like that is definitely one area that people are focusing on.
Matt, you have an questions here for Michael?
- No, I've know Michael for quite a number of years but I saw him give a presentation, and was really blown away, up at the Quant World Canada where we both froze.
It just was a breath of fresh air talking about the risks to short vol.
Talking about the fitness functions.
Giving a lot of advice on open-source machine learning sites to go for.
So I just wanted to have Michael just expand on some of those points that he gave at that talk he gave up in Canada.
- Sure, happy to do so.
In terms of the short vol points that we were hitting in Canada,
I'm not sure those are particularly relevant anymore.
That was all basically preparation for what happened last week.
- (laughs) we've all woken up to that last week, yeah.
- Yeah, so we can probably skip over that for now.
But if anyone has any questions,
I'm certainly happy to talk about vol as long as you like to.
But when it comes down to it, I think what's probably most interesting is when you think about options, the best thing to think about when you think about machine learning, are regression problems.
And these are all the hard MAP that we learned in probability and statistics.
Typically, a lot of investors, when they come to talk to us, anything they've heard about machine learning has been more what's in the general sphere.
It's not necessarily being related to a financial application.
So, for example, they'll wanna talk about image recognition.
Different things like that, that you would see at Google and you see more in the tech community because the tech community is obviously doing a lot better at machine learning than finance guys are.
But once you get over into the financial world, it's important to realize that a lot of what you're doing is really number crunching and coming up with the most efficient way to crunch those numbers.
And it's also really important to think about the instruments you're trafficking at.
And I think this is why it's probably great for your options podcast is options are a derivative of the underline.
You see a lot of people applying machine learning to the underline.
But that's a race to zero, kind of in the same way high-frequency trading is a race to zero.
Once you start to operate in the derivative sphere, there's so many mathematical relationships that have been modeled out on an academic level, but everybody knows that the academic understanding of options and finance doesn't necessarily translate to the real world.
- You made some points before as well in some of your talks.
And I've heard this brought up before, when people bring up machine learning.
The notions of supervised versus unsupervised learning.
And kind of some of the use cases.
And where some of these different types of learning are more applicable in certain types of trading and Alpha Generation versus others.
So, Michael, if you wouldn't mind, why don't you break your thoughts down on those for our listeners.
What we really mean by those terms supervised and unsupervised learning.
And then, where you feel they're more applicable in the markets.
- Sure, so it definitely, in terms of the applicability, it depends on what type of investing you're doing.
But, just to clarify for everybody, when we talk about unsupervised learning, we're talking about machine learning applications where you don't necessarily understand what's going on under the hood.
And this is actually one of the biggest push backs that you'll get from investors.
They'll take a look at, say, a deep learning algorithm or a deep learning strategy and they'll say, "Okay, I understand the computer
"tells you to do this,
"but you don't understand why the computer is telling you
"to do that."
And that's a very fair point for them to bring up.
And the manager has to be ready for something like that.
Conversely, once you go over to the supervised learning side, what you're doing there is you're understanding everything that's going into the recipe.
Every step that the machine learning program is taking, you understand it.
And you understand why it's doing it, because you're instructing it to do that.
Because you're answering a very specific question in that example.
So, for example, you might take a look at large data set of options and you could say, all right, if the market is doing X, how do all of these options behave?
With supervised learning you can go in and you can understand those relationships.
Whereas with unsupervised learning, it's more akin to going into the computer and saying, hey, find some relationships for me.
I don't really know what they're gonna be, but I'll trust you that these are applicable.
So both of them can provide Alpha, but the application of those is incredibly important to understand.
You need to know why you're doing what you're doing.
- Matt, I know you've been looking into a lot of this area as well, in particular.
Maybe some of the limitations of machine learning.
You have any questions along those lines?
- Yeah Mark, Michael, one of my pet passions is neuroscience.
And one of the guys that was on one of the podcasts
I listen to, was Jeff Hawkins.
He, way back in the day, founded the Palm.
You were probably too young to remember that.
But I remember it fondly.
But he talks about the limitations in machine learning and the terrific outlook it has.
And I tend to agree now that it should be based on a lot of the principles of the brain and become truly intelligent.
I think that's the future of it.
Do you have an opinion on that?
Or is that just me and my pie in the sky?
- There's certainly some limitations on machine learning.
And again, it's super important to go back to understanding why everything is behaving the way it behaves.
But it's interesting.
You might spend an entire career learning how to trade and somebody will come up to you and say, "Okay, well, tell me about your trading style.
"Tell me about your trading pieces.
"Why do you do what you do?"
And the reason that you do what you do, is because you've learned it over time from all of this exposure.
It's important to remember that with machine learning, computers are able to synthesize all of that exposure that you've had over an entire career and they can process it in a fraction of a second.
So you can have many, many careers worth of learning basically put into a certain algorithm or a certain trade.
And it's the most important part.
And I think this is where a lot of people screw up on machine learning.
If so many people are applying it to the Delta One space.
Meaning stocks or futures it either goes up or it goes down.
Where the real edge starts to come in, is when you start to deploy instruments like options and derivatives where you have capped loss.
Because there is some sort element of randomness to the markets.
There's certainly patterns that present themselves over and over and over again, but markets evolve.
And machine learning algorithms will evolve with the markets over time but it's important that you use instruments that are able to cap your risks so that when the market does evolve, and the computer's busy evolving trying to catch up with what the market is now doing, you've been able to sit there and manage your risks appropriately.
- Yeah, I couldn't agree more.
And I remember you talking about the fitness function up there.
And I also remember that you mentioned an open-source software that I can't recall it.
Can you comment on that?
- Sure, so, on the open-source it's OpenAI.
This is actually a non-profit that was founded by Elon Musk and a handful of other guys.
And what they do, is they put out problems for the world to basically digest and try and solve.
And these problems seem fairly benign these are not typically financial applications.
For example, I assume you guys remember the old video game Doom on the computer where you'd go around and shoot all of the aliens and werewolves.
You guys remember that? - Oh, great game.
Early foundation of my early (murmurs) into gaming.
(laughing) - For sure.
So what OpenAI is doing, is they have a competition called Vizdoom, V-I-Z doom.
And they'll give the guys this information and all the computer has are basically visuals of what's happening in the game.
And the computer has to break down those images and come up with an optimal strategy for the game.
Well, that's not terribly difficult to bridge from that over to the financial realm.
You can take visualizations, whether it's charts or whether you wanna encode your data through a picture in maybe a way that doesn't necessarily make sense to a human, but it makes sense to a computer, you can encode financial information into visual data and you can use a lot of the same applications and algorithms and processes that they're using in these OpenAI competitions and you can apply that to finance.
As long as you use the right instruments, meaning derivatives again, you're also able to manage your risk so that if you ever get your head blown off like you would in Doom, well you're losses are capped on your portfolio side.
- It's terrifying.
We're teaching the computers to kill already.
Michael what are we doing?
Haven't we seen Terminator?
We know that ends badly for all involved.
But I wanna go back to what you were talking about there a little bit, 'cause I think that's kind of interesting.
I've heard a lot of people talk about different aspects of machine learning recently and most of 'em, you're right, tend to focus on the Delta One side because it just seems like that's a lot easier starting place.
It's easier for the machines and the Algos to wrap their heads around a pretty much a binary product.
It goes up it goes down there's one stop.
That's kind of it.
It's a lot easier from an analysis perspective.
I haven't seen a lot of people really trying to leap into the weeds yet or saying that they're leaping into the weeds when it comes to the options and derivatives.
Typically, they view that as a little bit of a bridge too far.
They could analyze Google stock pretty easily.
We start getting into the hundreds if not thousands of options series and (murmurs) and strikes and calls and puts it can become an overwhelming bit of analysis for the Algos.
A lot of them maybe haven't lept and done that extra leap into the options.
But it sounds like you're saying that's actually where they should be starting.
That's where they should be focusing
'cause that's where the most Alpha, the most bang for the buck can be found for this in the options, is that correct?
- Sure, I don't wanna pound my fist too hard on that because that's what we're doing.
And that's how we're generating Alpha.
So I obviously don't need everybody coming over to the party.
But in the interest of transparency, yes I think there is a lot of edge there.
And Matt can certainly tell you this and I can tell you this that the fintech company I acquired, they had spent almost five or six years developing out data ingestion, ways to calculate the data, ways to monitor data integrity, cleaning the data.
Basically options data is so complicated because there's so many different variables that go into it, that unless you have a system that's able to go through and clean up that data and really make sure that you have high data integrity, it's really difficult to get into that whole sphere.
So once you have the systems and the applications are able to clean the data so that you can prevent a garbage in, garbage out problem, once you have that big data set, then you really have something you can sink your teeth into.
And again, I think a lot of people aren't doing this because the barrier to entry is high.
It's very difficult to go through and work through all of this data and ensure that you have data integrity.
Because when you go out and you just buy data from the data vendors, they're giving you data from the exchange.
I can tell you through what's been a very long painful process that I know this is true.
That the data that comes from the exchange, is not always accurate.
In fact, some of the data is so messy that is becomes very difficult to automate a lot of this BAP testing and a lot of this data crunching.
And if you don't have the ability to automate the data because you haven't cleaned it, you don't have anything.
And then, for most people, that's really to spend more effort than their able to put in.
- Yeah, you're right.
I often refer to that as the options fire hose.
This massive explosion of data coming at you.
The full OCC pipe, if you will.
You're right, it can be overwhelming for a lot of shops.
Particularly coming from the basic stock world where there's a handful of feeds handful of books and need to monitor, that's kind of about it.
A handful of dark pools.
Now you get into the options realm and it's just this explosion of data.
And you're right, it can be overwhelming.
Yet another reason why a lot of shops, maybe why Gammon Capital is lurking out there kind of alone in that space in terms of machine learning and the options front.
You know Matt, he brings up a good point.
If only I knew someone who had access to quality and clean options data and then I could go and perhaps run BAP test, I don't know.
You know anybody like that, Matt?
- Well, you see why I like Michael.
He's singing my song (laughs).
- Yeah, he's setting them up for you to knock 'em down.
I like that. - Yeah.
The exact same things that we talk about, Mark is this.
How long it takes.
And I'm happy to tell people how we BAP test, how we do all that.
But it is a chore to get the data.
And I use all my skill as a past market maker and clean the data and make our assumptions on the trades.
You could ruin a BAP test at many different steps.
Michael has taken care to, he uses an internal BAP tester of course.
OS presents a Client Facing BAP tester.
But, it's so great to hear Michael talk about this, feel someone else going through the pain and to have used a BAP test.
Michael, and once you get your results too, the fitness function I was impressed with in Canada about it's not just people looking for absolute returns, but it's also mitigating draw downs, mitigating volatility, et cetera.
Could you speak to the fitness function a bit?
- Sure, yeah.
I think this is something that's really important that a lot a people are missing.
And it's almost comical that since nobody really has a fitness function, it's made our conversations with investors far more difficult.
So, let me walk you through what a typical meeting with an allocator and potential hedge fund looks like.
The hedge fund will go in and the allocator will say,
"Okay, tell me about your strategy."
And the hedge fund will go in, they'll explain their strategy.
They'll pull out their track record and they say, "We do this."
And then, the allocator will either say, "Yes
"that works for me," or "No, that doesn't."
And typically, the hedge fund thinks about it in terms of how do I make the most money.
How can I maximize my return.
That's my ultimate end goal.
But, if you take it a step further, maybe you have a more sophisticated hedge fund and they say, "Okay, well, let me maximize
"our risk-adjusted returns."
Meaning a Sharpe or a Sortino.
Typically, we like to rely on Sortino just because we trade derivatives.
We have a lot of positive tail.
We are able to cap a lot of our tails, so all of our volatility's upside vol.
But the right way to have these discussions, at least when you start to deploy machine learning especially if you're doing this through a managed account or some sort of vehicle where you can really customize it to what the allocator needs, is you can go in and you can say to an allocator,
"What is the most important investing objectives you have?"
And he might say, "Okay, well I care 30%
"about win/loss ratio,
"I care 40% about risk-adjusted returns,
"and I care 30% about maximum draw down."
What we're able to do is after we've cleaned our option data and created these huge option data sets, we can take a look at the market, and basically put our data in context.
Are we in a high volatility market?
Are we in a low volatility market?
And then, once we've identified what type of market we're in, we can go through our large data set and we can say, okay, relative to what the allocator told me, what he cares about win/loss ratio, risk-adjusted returns, we can find the trades that are most optimal for exactly what his investing objectives are.
That's effectively what the fitness function is.
So we're able to create custom fitness functions for our clients and then from there, we're able to create strategies that best fit what they're looking for.
That is way more powerful than going in and having a hundred different meetings and hoping you stumble on the one guy who's interests line up with yours.
It makes so much more sense to be able to customize your offering to what investors are looking for.
And we think that's a spot where we really have a lot of edge over the competition.
- Yeah, like I said, I think that's great.
That's what we do as well with our OR statistics when we're done with the BAP tests.
Again, it's not off of just the annualized return but there's a lot of other things.
And you almost push into another portion of our BAP testing what we call TOES, tested option environment strategy, where we change the strategies based on volatility environments and stock environments.
So we'll make a matrix.
Are we in a high vol, high stock?
Are we in a high stock, low vol?
What of the nine, or of the six, I guess or whatever, and medium vol, what box are we in right now?
And what returns do best for various strategies?
Are you getting to that level where you're changing the way you trade based on an environment, Mike?
- Yeah, 100%.
When we look at the options, there is a handful of different inputs that go into how we define our environment.
Just as a quick peek under the hood.
Obviously, we care a lot about implied volatility in the market.
So, for example, the trades that we would have on a month ago, are now very different than the trades that we have on now.
We might also take a look at SKEW, see what that put/call implied volatility looks like.
We'll also take a look at metrics like days to expiration.
When you're sitting there and you're deciding which options are the best for accomplishing your investing goals and which ones best meet your fitness function, a lot of people will say, "Okay, well I wanna
"get long in SMP call."
Well, do you get long the one week call, or the one month call?
Do you get long the 20 Delta call, or the 80 Delta call?
Or do you get short or put?
There's so many different ways to express the same type of trade when you're trading derivatives.
So using that fitness function in conjunction with the different input variables to your model, you're really able to kind of hone in on exactly what you need to be doing.
We think that's really important.
- Yeah, it sounds great.
And obviously, watching for overfitting.
Many of these tail events don't happen very often.
And so, people will tend to overfit based on 2008.
Mark very much wants me to finish this project I have on an extremely long-term tail hedge (laughing) simulation.
How do you guard against overfitting?
And how do you work with the tail hedges given that we don't have very many events, Michael?
- Sure, so the easiest question there is how do you deal with the tails?
We are not net short options.
So that makes us different than, pretty much, everybody else out there.
Most volatility funds, they love to sell options and inevitably their day of reckoning comes like it did this past week.
This past week, that was not a draw down for us at all just because we manage our risks through these long options.
So that's really important right there.
And then in terms of the overfitting, there's a couple of different ways to think about that.
One, I've always been a large proponent of having product specialists using machine learning to get to where they're going.
That's a lot different than the majority of the hedge funds I've seen out there.
A lot of the hedge funds I've seen, they focus on bringing in the best machine learning guys.
They want guys who are doing cutting-edge machine learning coming up with new algorithms to apply.
But these guys might not understand the financial products at all.
So for example, they might run machine learning on a product like XIV and say, "This is amazing.
"Let's go buy a ton of XIV," without any understanding that, that product can get vaporized overnight because that's not in the data.
So one of the things that's super important that we do that I think a lot of people probably don't, is we put product specialists in the seat and they're the guys who are really driving the bus and they're using machine learning as a tool as opposed to being machine learning specialists who just happen to be trafficking in the world of finance.
So we think that's really important.
And then, another thing that we've been doing a little bit of R and D with that we're kind of excited about, is generating financial data that's never existed before.
So when you start thinking about Freckles and stuff like that, you can basically generate your own type of financial data.
From there, you can create incredibly large data sets of financial data.
And given the same types of metrics, maybe the same types of trends or so in this randomized financial data, we're also able to trade on that to help come up with a little bit of extra Alpha too.
That helps us avoid the problem of overfitting and kind of turning a blind eye to having highly fit models that don't necessarily work real well in the real world.
And then, the last thing I would say, is for a lot of the stuff we're doing, we're using supervised learning.
So we understand every step of the process and from that, we're able to minimize the overfitting because we're not turning a blind eye to any steps in our process.
- Yeah, I know you mentioned before the importance of testing your models on that out of sample data.
As you know, we've seen, as Matt kind of alluded to.
He's in the BAP testing space.
I'm sure he must get this request from clients all the time.
The people want the ideal BAP test for their scenario.
"Hey, can you create a BAP test for me
"that shows that my 5% (murmurs) put strategy."
Selling put strategy is the best thing since sliced bread.
Only in these years only in these timeframes.
And maybe you leave out last week and you leave out some other things.
And so, you selectively create a data set that you can BAP test on that makes your thing look fantastic.
And it sounds like you guys over there at Gammon kind of extol the opposite which is going beyond that.
Going out of your typical sample size and range to test your data on other sets to make sure that this stuff doesn't blow up in your face.
Is that correct?
- Yeah, you have to do that.
In fact, I would argue that anybody who hasn't done that and who presents themselves as having fully BAP tested and vetted an idea and a machine learning specialist who doesn't do any testing on out of sample data, those guys sound like they're begging for a lawsuit.
- Probably gonna see some of those unfolding after last week.
Something else you mentioned kind of caught my ear here which is, you're right, there is this,
I think it kind of depends where you come from in the machine learning space when it comes to the trading tech and fintech.
Which approach you take to this.
Seems like a lot of guys who come from, let's say, let's just lump 'em into the west coast kind of a Tech First kind of group who don't come from the markets.
They tend to have the view point or you kind of mentioned that you should have just an AI/machine learning specialist there and they could really do the brunt of the leg work.
But the AI algorithm kind of learn the products themselves and identify it's own and what the tech kind of solve everything.
Where it sounds like you're taking more, having come from the financial world, an approach on seeing.
I think a few more people are starting to champion.
For a long time it was the latter.
Now it seems to be, maybe leaning or pivoting a little bit more this way.
Maybe more after last week, we'll see.
You should have some sort of expertise on the team, in the product categories that your actually looking at.
For example, you're talking about adding you're own options component to a typical Delt One type strategy with machine learning.
To me, it would be fool hearty to not have someone on the team who at least understands the basics of the options mechanics.
Like you mentioned XIV, another important point.
Anything that can run a typical chart analysis over the past few years probably would have flagged XIV as a great trade without knowing any of the potential pitfalls associated with it.
So, I think you're right.
And in this day and age, it is kind of hard for me to understand how you could go down particularly if you take the leap into this space which is the options and derivatives space.
How you could do that without having at least someone on the staff who is familiar with the space.
Maybe you don't want me beating that drum too much though, Michael.
I don't know, 'cause it sounds like you guys kind of have this space to yourselves right now.
And so, maybe I shouldn't say too much.
- I put it out there, by all means, please.
It's definitely an uphill battle to be able to go out and educate investors not only on the volatility space, but also on the machine learning space.
Because both of those are pretty sophisticated arenas.
And you run into a lot of people who aren't familiar with them and it's easier for them to just say,
"No this really isn't a fit for me," than, "Go ahead and explore it."
So we're certainly happy to have the message out there.
But just to hammer home one point, and I think we kind of said it.
But it's just so important to remember that I really have to underscore it, is when you think about the hierarchy of competencies when it comes to finance, number one on that hierarchy has gotta be risk management.
Otherwise, nothing else matters.
And machine learning is important.
There are definitely great ways to use it to extract Alpha but if you don't look at the risk management component first, then it's just a matter of time until something goes south.
- Yeah, it is, again, as we saw last week.
Not to beat that drum too much but we saw last week that things can go south very quickly if you're not prepared.
As we saw some funds who won some major awards in the volatility space, not too long ago, vanish over the course of last week because they didn't have those proper risk-adjustments and risk management procedures in place.
I wanna touch on something else.
You mentioned, I heard it recently when I was down at the big Inside ETFs Conference mentioned a few times as well.
And I can't decide if I find this an interesting opportunity or completely terrifying.
Maybe the correct response is a little bit of both.
It's this factoid that 90% of the world's data was created in the past two years.
On one hand, I can see how a data scientist would be licking his chops and saying,
"Wow, that's a fantastic opportunity.
"Look at how much data I could mine
"and correlations I could find."
And from the machine learning aspect and everything else there's a lot of potential opportunity there.
But I also look at that and say, "Wow, we've created
"a lot of nonsense and a lot of waste."
A lot of digital waste and noise over the past two years.
More so than, perhaps, in the entire history of human civilization combined.
Which, perhaps, does not bode well for the future.
So, I don't know.
Maybe, Michael, you come at it from a different perspective coming out of maybe a data scientist perspective.
Maybe you find that tremendously intriguing or are you maybe a little bit concerned about that like I am?
- It's certainly interesting.
And in time, we'll have enough data to be able to do everything.
I couldn't be more excited to be alive right now.
Of all the times in history to be alive, this is, hands down, the best.
But what you just said, really brings me back to the point that
I try to hammer home with people, is it's so important to have product specialists in the seat because they can understand what's going on with what the algorithms are telling them.
So that's really, really important.
And then, in addition to that, it's super important that you do a fair amount of supervised learning as well.
Especially if that's one of your core competencies is a certain product.
The supervised learning really makes everything that you're pulling into the model relevant.
So you're not pulling in a lot of noise.
So, for example, let's say I wanted to make a bet on SMP going up or down.
And we have this giant data set of everything all over the world.
And we're looking at some monsoon in India or something like that and there happens to be some correlation over the past two years.
Well, as a product specialist running a supervised process,
I can tell you that, that's garbage.
It doesn't make sense to me that the weather in India is gonna have any sort of reflection on the SMP.
Even if there's a high correlation there.
You gotta remember, correlation does not necessarily imply causation.
So that's why it's so important to have these product specialists in the seat so they can really understand what's going on in their algorithms.
- Well Michael, it's been a fascinating conversation.
I wish we could keep going.
We have to get onto, actually, you've gone so long, we don't have time for Hot Topics.
We'll get right into some listener questions in a bit.
But before we do that, really quickly let our listeners know if they're intrigued by what you guys are doing over there at Gammon Capital or perhaps they wanna reach out to you with some questions on machine learning, where should they go, what should they do?
- Yeah, I would say, first off would be the website, www.gammoncap like Backgammon.
We've got a history of Backgammon in the firm.
So Gammon Cap is a great place to go.
That's where people can start to understand what we're doing.
Especially on our insights tab.
You'll get a look into the way we think about the world.
The speech that I also gave at the Quant World that Matt had referred to, that's on the website.
So if there's any questions, start at gammoncap.com.
And if anybody wants to reach out, they can just reach out to contact@gammoncap or ir@gammoncap and we'll be sure to help continue the discussion.
- All right, I know you guys have been waiting patiently out there for us to tackle your legion of trade and technology questions.
So we're gonna do a little bit right now with our special session of the Inbox.
- It's time for you to take your place on the Trading Tech Talk panel.
It's time to open, the Inbox.
- All right, we got a lot of questions and comments let's get to 'em.
Actually, one 'em kind of came in right as we were about to do the show.
They heard that we were gonna talk about AI and machine learning on the show today and they chimed in.
This is from Mark, Mark Gr-a-nt.
Listens to a lot of our programs on the network.
He wrote in with a comment, which actually is kind of apropos to what we were just talking about, Michael.
He says, "If AI is so smart,
"why can't it trade options for a profit?"
It sounds like you're saying that, that's exactly what it can do, Michael.
- I would say it can and has.
I think the reason he's probably asking the question now, is the AI space, and especially the options data space, is so complex that most people just haven't tried to tackle it yet.
Even some of the most sophisticated multi-billion dollar firms out there haven't tried to touch it because the data crunching is just so difficult.
Like I had mentioned, when we acquired the fintech that had been doing this, they had been working on that project for about five, six years.
I know Matt has certainly got fair amount of time in on his IP as well.
So I think the fact that there are so few players in the space, is probably what leads to a question like that.
But to be fair, there are some of us out there that are doing well with that.
- Yeah, I guess your existence is proof.
At least you're the exception that proves the rule, at the very least.
'Cause you're right, the perception is out there that the AI, the machine learning, call it what you will, hasn't penetrated to a large degree yet in the derivatives.
There's still a little bit of a bridge too far.
But we're starting to see end roads, particularly in beyond the Delta One type futures and really into the options.
And that sounds like we are starting to bridge that gap there.
Let's see here.
Let's go, since we have you on, Michael we'll focus on some of these
AI and machine learning type questions.
We got a lot of those as well.
Let's see, Bloud9 wants to know, this is kind of what we touched on earlier.
He or she asking, "What do you see
"as the outlook
"for AI and machine learning
"in the financial management space going forward?
"I see a lot of smoke, but is there much fire?
"Are there many funds
"currently using it to generate Alpha?"
All right now, well, Michael, we just discussed this about be fresh in your mind.
But kind of reiterate for our listener here what we kind of just talked about earlier.
Obviously, you guys are out there doing it.
Are you seeing a lot of funds really using it to generate Alpha right now?
Or is it still, very much, in the testing phase for most of the funds?
- With respect to the options,
I have yet to come across another firm that's doing it in a derivative space.
So I can't really speak to other guys out there that are doing that.
Obviously, in the Delta One space, you're seeing a lot of people making moves here.
Some people are doing it better than others.
But, I think the way to think about this is the way that you think about any sort of technology.
And that is, does it provide efficiency gains?
If a brand new technology provides efficiency gains, that means that there's something that happened there.
It's gonna get adopted in time.
It's gonna get used people are gonna learn it.
Some people are gonna make mistakes as they're learning.
Again, that's why it's important to do it with options because if you make a mistake, your risk is capped.
But overtime, you'll see AI and machine learning take on a bigger and bigger weight in the market.
I think that's definitely gonna come.
I don't think there's any way to avoid it.
And anyone that's turning a blind eye to it, I would hope they're at the tail end of their career because in 10 years, you're gonna be needing to do this or you're gonna be out of luck.
- Anything on this, Matt, for our listeners out there who wants to know any funds?
Are you seeing, when you're out there engaging with these funds who are contacting you for BAP testing and things, are you seeing a lot of funds that are really using these approaches to generate Alpha?
Or is it still very much they just want the data to kind of keep crunching the numbers and testing their systems before they really commit capital (murmurs)?
- We have some of the largest hedge funds and I know they're doing something like this.
They're consuming every bit of data that we can get out to them, every new bit of data.
And they're tight lipped, obviously, about what they're doing.
But they're paying and they're getting a lot of the data that we've been making now since 2001.
So they've gotta be doing something along those lines.
They're just not talking too much about it, Mark.
- Not surprised, hedge funds are tight lipped.
Who would've thunk it?
All right, we got time for a few more here.
Ton-ic, To-nc, however you pronounce his handle, wants to know,
"What are the best ways to invest in blockchain right now?"
Well that's obviously the hot topic.
I didn't even have a chance to get to it on the show today which is kind of a relief (laughs).
(mumrmurs) stop watching all the time so it's a little bit of a respite from it.
It's not the worse thing.
That said, there are a lot of ways people do this.
In fact, obviously, you can go he direct investing angel/VC route if you wanna get really, with your risk capital and invest in the legion, there are infinite number of blockchain oriented startups out there.
Pick the industry, there are trucking, freight.
The adult industry now is getting into blockchain.
So maybe that's a pure sign that the bubble has peaked.
So pick an industry.
There are firms using it if that's what you want.
If you want a specific flavor of blockchain, it's probably out there for you.
But I actually, ironically, I just came back from, like I said, that big Inside ETFs Conference.
And one of the big talk down there was that someone had launched the first official "blockchain" ETF.
I think (murmurs) was actually just blok, B-L-O-K.
And so, I went and looked at the holdings of this first blockchain ETF to see what it is that they hold, and the lion's share of the holdings in there are names you can easily go out and buy right now.
Amazon was a large holding.
Microsoft, a couple of other large tech names you probably already have in your portfolio were substantial portions of this ETFs holdings.
So you may very well be invested in blockchain
"right now" and not even know it.
Granted, it's a small part of what those firms are doing but that was a large lion's share of the holdings of that ETF, at least at that time.
So, if you're investing in those names, you could already be investing in it but you could have created your own little basket of blockchain and not even known it.
So it's easier to do than perhaps you might think.
Michael or Matt, any thought on this?
Obviously, it's the hot topic.
It's topic di-gi-or these days.
Everyone wants to touch blockchain.
Any thoughts about how they should go about it?
- Yeah, I would say, in terms of blockchain, you see a lot of people who are just getting into the space and they don't necessarily understand the difference between a lot of these currencies.
So they might take a look at Bitcoin and think that's the same thing as Ether.
Well, that's not really the case.
Ether, you're actually buying processing power on a decentralized computer.
There's actually value there.
There's something behind your money other than the whole faith in credit of all the other users.
So, there's big differences in the different types of cryptos out there.
So for people that are looking to invest directly into crypto,
I would say if this is a hobby of theirs or something that they're interested in, by all means, go ahead and do the research.
But, if you just kind of wanna dip a toe in the water, it might be a little bit better to hand your money over to some of the professionals.
I know that some of the blockchain hedge funds out there, some of them can be pretty expensive.
I also know that the minimums can be kind of high.
I actually just heard about crypto fund of funds that came out that has super low fees.
The Bitcoin investment group run by a guy named Eric White.
I know he's been having some success out there.
He's able to take smaller tickets and he can aggregate them and he can deliver the money and actually get a discount on some of his fees to some of the more sophisticated players out there.
And those are the guys who really understand the differences in these technologies.
They can tell you, "Oh, this technology
"doesn't have any real value behind it.
"Whereas this technology, you're actually
"buying something other than a holding
"for your money.
"You're actually buying processing power.
"And this is the way it works.
"And this is the way the industry is moving."
So, Eric's situation over at Bitcoin Investment Group really gives you a chance to kind of hand your money over to the experts and let them focus on it.
- There you go.
Let the experts focus on it.
I was surprised very much by that ETF.
Kind of just how broad the whole thing really Seemed.
In spite the name, it didn't really seem as blockchain focused as you might expect from that.
Let's see, we got a couple of more we can squeeze in here.
This kind of again goes back to what we were talking about before.
Mark Ca-lus asking, "How are the computers
"getting the edge on us in trading?"
He thinks, "Is it just pure correlation trading?"
I think, Mark, we've kind of talked about that already a lot on the show here.
That they're, at least what Michael was talking about here, about using the options, they are taking that next stop.
I think that the correlation stuff was probably the very early the very low-hanging fruit out there in the AI machine learning.
What happens to the VICs when the SMP goes up and down and vice versa?
What happens to gold?
I think those are probably low-hanging fruit trades.
Maybe I'm speaking out of turn.
Michael, is there a lot of easy correlation stuff left?
Or is most of that low-hanging fruit pretty much been gobbled up at this point?
- I would say in the Delta One space, that a lot of that's probably been gobbled up.
But you also have to remember there's new information coming in that you can process every day.
So the information that people decide to (murmurs) in their investment process can really distinguish one investor from another.
So, there's always a little bit of edge there.
And basically, being the best analyst and coming up with the most relevant data.
Having said that, I'd also go back to pounding my fist on saying that using options is so important.
Again, because the data sets are so sophisticated, that a lot of the correlation trades that are out there, those still exist in the options world and those are able to be taken advantage of if you have the proper processing power to do so.
- Matt, any thoughts for our listener here, Mr. Ca-lus?
He wants to know is it just pure correlation or are they moved beyond, so it sounds like they've moved well beyond that at this point.
- But I would echo what Michael said.
There are still opportunities in the options space to find these correlations for volatility.
And like Michael said before, for the skew or the slope as we call it.
So there are correlations to be found in the options space.
And a lot of the tools and a lot of the data sets that have worked in the past for Delta One can work for options.
And the new data sets are being tested for options.
There's just a myriad of more opportunity in the options space because you can structure the move for a particular strategy to mimic the data.
And I would say, that the computers can see these relationships.
And when you start talking about all the different relationships and options, that's where the edge starts coming in from this AI machine learning mimicking the brain for pattern finding, et cetera, Mark.
- All right, we got time for, let's see.
See if we can squeeze in a couple of more people that have been waiting so patiently for our next episode of Trading Tech Talk
I hate to leave them off.
I'm not gonna fit anywhere close to as many as we can.
Even as it is, let's get to as many as we can here.
I got a comment here from @bon-skew, I like the handle.
That's just a good handle, bon-skew.
So we gotta squeeze him on.
He's actually responding to what we talked about on our last show.
Which at the time we asked, "Are exchanges really
"shooting themselves in the foot
"buy continually hiking market data fees?"
We were debating that at the time on our last show.
He chimed in to say, "There's no real debate to be had.
"There is a simple and foregone conclusion.
"Higher fees equal lower participation."
He adds a little caveat a little addendum there at the end.
"Perhaps a goal for Central Book market manipulation."
I don't know about that.
But yeah, he agrees with the fact that,
I think everyone pretty much does at this point, outside of the exchange space, that the ever increasing reliance on market data fees to make that bottom line is probably squeezing the markets in ways that people did not want or certainly didn't anticipate going into it.
And the last up, we got here, that's all we got time for, is C, C16 asking,
"Can you please do a deep dive
"into smart routers
"on a future program?
"They're the secret sauce of the market these days,
"but no one really seems to understand
"how they work.
"Shedding some light on the otherwise dark world
"of routing would be a great service."
Yeah, I'm with ya there C16.
I think that would be a fascinating program.
We've talked about it behind the scenes a few times.
Maybe it's time to really get that show together.
'Cause maybe that would be fascinating.
Particularly for our neck of the woods, options.
Routing is pretty much required at this point, to have a smart router to be able to really play effectively in this space.
Whether you're market making, you're having a seat at that table or you're on the brokerage side, certainly routing your flow to the best venue and outlet is crucial to your success as a firm.
So maybe it might make sense to put together an interesting little deep dive into how those routing decisions are made in 2018.
I think that would be a fascinating thing to discuss her on the old show.
But that music, unfortunately, means we've come to the end of another fascinating journey through the world of trading technology.
We got to as many of your questions as we can.
We'll get to more on the next show.
And of course, we had nice deep dive into all things machine learning.
Particularly on the options front which is an area that really hasn't been discussed too often in the past.
Michael, I'm glad you could join us.
Before we go, you mentioned where people could go to learn more.
Maybe just hit that one more time for people that wanna learn more about what you guys are doing and where they should go.
And also, maybe if you wanna leave our audience with any hints, any teases of what's coming down the pike from you guys over there at Gammon Capital, now is the time, sir.
The floor is yours.
- All right, I appreciate it.
Yeah, anybody that wants to take a closer look come on over to www.gammon, G-A-M-M-O-N cap, gammoncap.com.
In terms of what's coming down the pike, we're just gonna keep building out the models.
Keep synthesizing more data.
We're surrounded by a ton of low-hanging fruit so there's a lot of opportunity right now.
I would say, anyone that's interested in learning about machine learning, especially on the options side, it's a great time to come out and have a chat and see what that world looks like.
- There you go, reach out.
Gammoncap.com to learn more.
And Matt, same question for you.
What is coming down the pike?
What can we expect from you guys?
You guys have certainly been doing yeoman's work out there in the earnings straddle studies of late.
So I'm glad to see that.
And of course, what you just put out about the bids and the slippage over last weeks trading, certainly fascinating stuff as well.
What else can we expect from you guys over there at ORATS in the coming months?
Uh oh, gotta find the little red mute button, sir.
- You're right. - There it is.
You pulled an Andrew.
Someone had to do it. (laughter)
- As people start to use and find strategies using our BAP tester, it was interesting to hear the question about smart routers
'cause we're getting questions about how to implement the strategies and using algos and how to leg into certain spreads and such.
So we've been working on that and forward testing our BAP tests.
And like you said, it's earning season so we have a plethora of earnings data that we're sending over to you over at Option Insider.
And yeah, looking back at last week just to try to answer some of our client's comments on what happened to the market, we put a nice blog post out on how the market's widened and liquidity kind of dried up there, so.
That's what we're up to over at orats.com.
And you could contact me directly
- There you go, check it out.
Orats.com to learn more.
And on behalf of Matt and Michael and indeed myself,
I thank all of you out there and the listening audience for downloading, streaming, subscribing to the show.
Of course, for sending in so many different questions.
We love to hear from all of ya.
And we'll see you next time, for more, Trading Tech Talk.
- Thank you for listening to Trading Tech Talk.
You can find more episodes of this program along with detailed show notes at theoptionsinsider.com/ttt.
While you're there, be sure to post a comment for your chance to be featured on the next Inbox segment.
Trading Tech Talk was brought to you by Trade Alert LLC, the market leader in Order Flow Analysis.
From Wall Street to Main Street visit tradealert.com to see how our specialized tools can help you master your markets.
- The preceding program was a production of The Options Insider Radio Network.
For more quality Options programs, visit www.theoptionsinsider.com or search for Options Insider Radio Network in your podcast provider of choice.
Listeners can also access all of our programming through our mobile app available in iTunes and on Google Play.
Don't forget to follow along with your favorite programs and submit your own questions for the host