クオンツトレーダーとは？ (What is a quant trader?)

字幕表動画を再生する

Hi everybody, this is Shaun Overton with OneStepRemoved.com
and in this video I have Michael Halls-Moore. Michael's the owner
of quantstart.com, a blog about algorithmic trading. Mike thanks for
joining me
Mike, you have one more interesting stories about how you got
involved in finance
You want to tell us a little bit about it? Yeah, I came out of
postgraduate school and essentially, one of my friends actually
after going through a start up
He actually said would you like to be involved in an investment of his
which was essentially a quant fund, a prop fund A nice guy to know.
He's a nice guy to know, exactly.
I became essentially they're systems developer or quant developer.
And then
with that we essentially traded for about two months
we really started from scratch. Everybody has this
impression of what a quant it is. You're a quant developer or quant trader
What is that? The quant word is overused quite a lot
what I think what people really means is something that involves generally maths or
statistics in finance and in a quite heavy
fashion. It can mean anything from
the guy programming the trading structure right through to the
the guy doing the hardcore trading research
Strategic back testing in a very
computational and statistical manner Right
it's very nerdy. You got the propeller heads. They're in the back room
Exactly. When you're developing a strategy, how
important is the actual rule for buying and selling?
Not as important probably as
other aspects. I mean, you can think of it as one big whole
system, right? You got you got pricing to deal with, data cleansing you've got the
what they call the Alpha signal generator right that you mentioned
and then you have
the whole execution and an exchange model and an execution process and
each those areas has its own
unique challenges essentially. I see that sorta
signal generators
are one bit of a much larger system, which is not as
The importance is on all of these systems working absolutely well
rather than one signal. I get to use to signals
not to highlight any system, but say have a moving average cross. something everybody
knows
You look at the signal and you think, "Oh, if I would coulda
should have done this then I would have make loads money," but
that's not what funds or quants do at all.
It depends on the frequency of what they're trading, which is a
very important factor. It probably differs quite a lot from retail. Retail
is very much at the lower end of the frequency.
There's still money to be made there of course, right?
Quant funds tend to be generally more of the higher frequency. A lot of
what they will be studying is optimizing of technology and
strategies that are really involved in market
microstructures and like say of the
particular assets they're trading. That is quite different approach
generally to
"You know, I've seen that. I could've bought there. I might've sold there." It's a very different sort of
approach. The signals are generated in quite different ways, I think.
What is the typical timeframe? What's the difference between a retail
trader
a guy that just has a trade robot that he bought off the internet and
a guy that's developed it and he has a statistics background and he is a true
quant?
Depending on what you're trading it could be anywhere from
the guy who has bought the robot
That sort of guy would want to be
checking himself making sure it's not doing anything stupid. And so, you really wouldn't
want it to be doing
more than a few trades every minute probably.
That is the absolute cap? It's whatever you can
cognitively handle. In the true quant sense
the ultra high-frequency stuff will be microsecond and even nanosecond
kind of scale
that obviously is not, you know, it takes three hundred milliseconds to
blink.
You can imagine the kind of latency that
you have to deal with in that space is a very different kind of mentality. It's all done with
a computer. When you're competing against
quants that are trading in microseconds and
your technology is cut streaming from the US to your broker server in the UK
or whatever
how do you compete? You essentially don't. You are playing on
a different playing field, I mean
They're dealing with
differences on these incredibly tiny time scales. They have their own little
games going on down to that level and
As a kind of longer-term, lower frequency trader, you're playing
You aren't really hugely sensitive to latency. If you're trading
once every week, latency is not gonna be your biggest
problem Having been on the inside and worked in funds
where do you think the retail trader has the biggest opportunity?
Ok, so there's
as a retail trader, you could go for more a higher frequency style
straetgy and there are plenty of
spaces out there where bigger funds, because they got quite a lot to
invest
aren't able to compete at the lower scale of the money.
As a retail trader, you can go there.
You can make money even with a relatively sophisticated
strategy. But you do need to be quite computationally
aware. You need to be good at programming and probably have a bit of a
background in math and stats to compete at that end.
When we talk about quants, for whatever reason, that is almost always associated
with high frequency trading or even ultra high frequency trading
in my mind anyway.
There are plenty of quant funds who tackle lower frequency stuff.
There's quantitative and there's
automated. You can have a very low frequency, automated strategy.
It's just that quant funds by virtue of generally who they hire
they'll be working at the higher end of the
spectrum because that's generally where they have the skills to compete.
Why is that? Why do all the funds focus on these short term strategies. Get the
money
right now, real quick. I think it's because they take
quite a scientific approach to how the trade so they'll be actually trying to
exploit
physical inefficiency in the system, some of the time. It won't be
quite different from, as you say, coming up with a technical signal and going for it.
They'll be saying, right, hang on
the exchange works like this, they use this infrastructure
Can we exploit that somehow? So, it's just pure arbitrage?
Or something very close to it?
There's lots of different strategies, one of them obviously being arbitrage
There's lots of we know the infrastructure better than the other guy.
It is done that way. You can see how it comes from
very much a sort of hypothesis first
trade signal second mentality rather than, "Oh, let's
see what indicator gives me a buy or sell." It's done in a very differnt
way
What is your opinion? Do you think you need
the science background first and then you try to figure out trading or is it
better to have the trading background and then you try to develop a strategy
I think you need a mix of both.
You can't just be a nerd and have a Ph.D.
No. You can come up with the best
forecasting system in the world.
If you don't understand risk management, if you don't understand position sizing
You don't understand these core
bread and butter ideas in trading, you're going to lose money.
You won't have done any statistical analysis on your
drawdown characteristics.
Or even know what drawdown is. Or even know what drawdown is.
You cannot, I think
I believe you need to have a mix of both, frankly, to do well. Before we
started recording, we talked a lot about risk management.
What is that in your mind and how do you best go about coming up with the best
risk-averse strategy that makes money?
I think retail and funds have a very different approach to how they deal with risk.
In retail from my experience it seems kind of a secondary
consideration to the the signal generator. In a bad way or good way?
In a bad way.
It seems like a dull, boring area of
trading. You can see why that is.
You're telling me I can't make as much money as I possibly can. I have to dial it
back. Exactly.
This really comes down to the differences in now retailers and
funds operate. Funds have institutional mandates
What is that? That is, they have to generally
present monthly performance for what they're doing. Some investors will actually ask for
daily
performance or even real time performance. You're really under
the magnifying glass a lot of the time. For that reason, the primary concern is
not losing money rather than gung-ho. This is for the institutions?
Yeah, this is for the institutions. Especially for your larger institution like pension
funds who the people are putting in Well, it's people's pensions. It's people's
pensions. They're not going to just throw
a hundred-million at you and say, "Go for it."
They will do an extensive amount of due diligence on you. Years of due diligence in some
cases. They will want to see very sophisticated
risk systems in that regard. I guess a few examples would be
They'd want to see exactly
how you go about leveraging up and down and what your actual
mentality is for leveraging up and down. It's not just picking it out of the air.
There's a protocol? Very much a
protocol and you know it when I see it I downside insurance policy would be
in that regard. A classic
put option to make sure your portfolio.
They're stress testing you.
Yeah. You'll also be stress testing yourself. Yeah, of course. All the time.
There's the industry standard like VaR, the classic
stress testing. It's more a culture difference I think. It's
a risk
first mentality. You're always thinking how much is this going to
affect my portfolio swing
How much of mine is going to be the average volume
that I'm trading. It's just constantly assessing risk.
Whereas in retail, it's mostly, "What's the signal, what's the signal?"
exactly so if you are retail and
obviously retail traders are primarily profit-driven
How does that change your approach? If you were in an institutional environment
your risk management focused in terms of you want to make money every day, even if
it's a fraction of the potential money
But if you're trading your own money, within certain
restrictions and you can only tolerate so much drawdown
within that drawdown. you wanna make as much money as possible. How do you go
about doing it?
There's a couple of actual, specific ways. The classic is
the Kelly
criterion, which is a means of adjusting your leverage based on the amount of
account equity you have.
It can get a bit mathematical so I won't delve too deeply into that.
You can Google it.
The more conservative uses of
leverage and cutting back on leverage. That would be very much controlling your
drawdown and your
growth rate as a sort of slide almost. You say, "Alright, well I'm happy to
reduce my drawdown at the expense of not gaining as much growth."
Obviously, any trader doesn't want to lose their entire equity.
That's called gambling. Exactly.
They'll all be fundamentally using some principle like
Kelly
It's quite a common one. You can go anywhere right up to
as I said, the put option mentality
basically only half your account and then literally put 50%
of your account aside
if you lose your entire amount, you've only lost
50%. That in itself is actually quite a
good mentality to have in some regard. If you... I've used Kelly before
and sometimes you have systems that have done way too well in the past
and you plug in the numbers and the optimal Kelly is something like risk 30 percent
of your account on some
but I know from experience that's dumb. What do you do? How do you
handle that? I mean
You tend to have overlays on top of that.
Another actual mechanism that is used
a lot in quant funds is what's called leading risk indicators.
especially in equities where you've got these new shiny
exchange-traded funds popping up all over the place. Some of them will be
measuring
very interesting things like the VIX, which you've probably heard of
This is the implied volatility index of the options on the S&P 500
and some people use that as a mechanism for essentially saying what is the future
volatility goign to be looking like. You would incorporate a lot of these.
If you can imagine putting all of these different factors into a pot
and essentially saying well okay my overall leverage is gonna be
this, it's not just dominated by Kelly. It's not just dominated by the VIX.
Mixing them all together and getting a holistic view.
You're mixing. You're not just saying here is my entry signal.
Here's my system that goes within
the entry system to come up with
the best trading size.
The signal generator is piping information to
a sort of more important risk management system that says
okay yes, okay no, or downscale by this much
the signal generator is convincing the risk manager whether or not to do the
trade rather than
the signal generator being the be all and end all decision-maker
the risk management layer is generally much more important and much heavier
Do you think you could make money if
say that you're trading Forex or equities or whatever
and you're pushing through a hundred shares every trade and you have stop-loss of
fifty cents
every time. Do you think you could make money doing that?
Essentially, because stop losses
they are quite a controversial thing to do in
trading. Are they? Cause retail traders... In trend following
definitely, they are a fundamental part of the strategy
You have lots of little
mini losses and then all of a sudden. The mega-winner.
Obviously, if that reverses, you don't want to lose it. That's the whole point of a stop
loss, but in
mean reversion, they're a bit more difficult to utilize.
Even a lot of quant funds specialize in mean reversion
they may or may not be using. So you have these huge institutions
that are really prestigious and
they're not trading with stop losses? I think you'd assess it in a
different way/
You'd say to yourself, right, what is the kind of historical
distribution of returns we've seen in our backtest.
and then you'd say well I've seen a 50 percent drawdown
in the backtest. You might just say, let's get rid of the
backtest and start again.
but you you'll see some drawdown in the back testing. You say ok, well
in the future I'm likely to see that drawdown if not much worse
You would essentially cap it in that regard.
That brings up a good point because the signal works
and you back test it. You feel good. Unless you get really
unlucky, it should work up to some point in the future
and then it stops you so that begs the question
how do you know when an algorithm is in a drawdown
and when you need to turn it off?
Before you even begin trading, you've carried out the backtest phase.
You would then say, right, I've seen a 30 or 25 per drawdown
in my backtest, which to me would be pretty high.
That's hard to ride through. Let's just pick a figure - 25%. if I got
anywhere near 25 percent in my
in my walk forwards, in my actual out of sample test, I'd just turn it off.
Time to do some different?
The reason I would turn it off is not because I'm expecting a 25 per cent in
the future it's because
I'm aware that I do not know. There's this unknown unknowns that could creep in and
it may even get much worse than that.
You should be expecting worse drawdowns during the walk forward.
yes a you might you should it be expecting were stored and certainly in
That raises a good philosophical point.
You can't know that your strategy is going to make money in the future.
How important do you think it is to try to disprove the strategy?
what's the value in that? I would actually come at it from the point
of disproving it. I would use the backtest not as a means of
putting something into production. I would use the backtest as a means to filter stuff.
getting... imagine starting with 100 strategies.
You can then use a computer to backtest the lot. You would filter 95% of
them out immediately.
Just cause of drawdowns?
Drawdowns, you're not happy with the statistical properties
For instance, you might not like the fat tails
as they would call them. It might be too much for you.
You're always looking for a filtration mechanism. You want
to find a reason to get rid of it always. If you
If you're happy that you've not found
any reason get rid of it then you'd put it in, as opposed to I've start with this
let's try and find everything I can to keep it. Very different
mentality
You mentioned fat tails and we've talked about trends
we talked about range trading. What's the relationship between those trading styles
and the fat tails? What are fat tails for people that don't know. Fat tail would mean
A lot of the finance is based on the assumption that
returns aee normally distributed meaning
they have relatively thin tales. There's relatively less
extremes occurring then probably do in real life
You'd say they have fat tails. This is the kind of
Taleb approach. The Black Swan guy? The Black Swan, right.
there's more extreme events than would occur
than you would assume under a normal distribution. You would be
using a distribution that has
much more extreme events more commonly. In the quant world
because Taleb writes scientific papers and he writes Black Swan in pop
literature type stuff
how how popular is he? He's a character, certainly.
He has very interesting strategies that sort of exploit
his belief that these fat tails occur more often than not. They are difficult
strategies to carry out. How are they difficult?
They are the opposite of what I would call the pennies under a steamroller
approach.
You are taking, it's very similar to a
momentum or trending strategy where you take lots of mini losses
on selling these options, out of the money options and then
every now and then, an extreme event will occur and these options will come into the money.
He believes these options are mispriced such that
you can make more money. Do you ever know when
you're due for a winner?
No. You can only ever say
historically over this time period I've seen this amount and
you know you can never be certain. You're trying it always
to see if it fits the statistical distribution. You don't know what that distribution is in
real life
You can only sort of estimate as best you can that
distribution and then
hope for the best, right, in his case. In my opinion it's quite
a tricky strategy. If you, let's say that you have a suite of range trading
strategies and have a suite of trend trading strategies
What do you feel more comfortable with in in the future?
This is a discussion that goes
very much goes back and forth depending on which year we're in, right?
Where am I making money? Exactly.
I would say quants tend to be... some quants are
ambivalent. They will basically try and assess
what type of market regime they're in and then apply
Do they do that quantitatively? They do.
I mean this is quite a sophisticated area actually and it's a very tricky one to get right.
There's a lot of them filtering mechanisms that
kind of machine learning technology that would be used to
try and ascertain whether you're in a
range-bound or mean reverting
scenario or a trending market. There are lots of ways of doing that but
you gotta say, "What time scale am I talking about? Am I talking about years? Am I talking about days?"
When you have a market regime shift
I mean
it's an unknown unknown
in some sense you can have a kind of view on interest rates
You can model interest rates, but
that's kinda a purely random event in some sense. It's very
difficult to
keep it in mind. You'd have to kind of look over historically what the ECB had done
and say how likely are they to drop rates?
The more sophisticated firms will obviously be tracking all this fundamental data and
be putting it into their models.
In some sense they would have more of an edge
in that regard to be able to detect these things, but
It's tricky. Market regime detection is one of the hardest problems to
solve in quant finance.
for that reason, to actually be able to answer whether we're mean reverting or
trending strategy is better at any time is very asset dependent and very
dependant on which year you're in. It's a tricky one.
With trading, it depends on what you're trading. With forex
there's like twenty to choose from. But if you're trading equities
you have the opposite problem. There's five thousand to chose from. How do you
pick one?
I wouldn't be
going and picking stocks in the traditional sense. I would
try and find some kind of economic relationship between
a pair of socks or a triplet of stocks in ETFs generally because you can
take a more... Yeah, why are ETFs so popular?
They're really quite interesting because what they allow you to do is
a company, the fund that sets them up can basically say, "We're doing
this specific thing and we will then give you an exchange-traded
instrument that does this so you might have a a possible gold stocks right now
i'll settle at war you might have
any shifts the triple levels up the nasdaq fortune minus
triple letters the nasdaq or something like that and say time
you've got low sophistication to buy and sell the system is without having see
and because they will exchange traded a base in stop-start
celiacs time and see deborah may also use generally
very very liquid the on on certain things saying I'm
sorry that take okay it might be the biggest example I can think up with an
e.t.a
and tracking error is USO yes the oil
ETF in the US market and I don't know what it is now but years ago it was 10
percent two years enormous
are ETF's the dumb money for you know that they're gonna do something in you
exploit it when they first came out there was a heavy load index of charge
not try to
United special night live. the classic S&P 500 and you know the
St line 119 keep trying to exploit differences in that and
here that can still be done you're you're investing significant man you
technology to do that you
okay so what is the advantage in the current market with the TSC
I would say it's not so much the any kind of trolls on on that so much for
the retailer Lisa for talking
so it's stability say well like a hand
this ETF has some peace of mind this is something to do
oil and there's fucking a fundamental underlying relationship between these
two quantities
okay and the you you could come up with a clutch scientific hypothesis
for why say a company that has a basket gold stocks
plus another easy if that measures the spot price goal tracks as well for
school
put in some sense because places or you know and you can exploit high
correlation
and and try to fix excuses is a much more scientific approach
to you here you know basically take a punt on any firm
you basically trying to exploit some kind is economic relationship that
should
partnership the definition BTS stay there
okay so you aren't just running math models you do have know what the market
is and FBI mean YouTube the
euro is coming with economy's fundamental idea is a your I
it they're all she complains that he really questionable may just literally
from focus prices from prior prices without any
concern about I'm fundamentals i mean
my personal views you need to mix the two okay because these
Black Swan defense right i mean if if there's no you
prior horrific volatility in the prices reason new
base model not been you know your PC is a
you not to expect a pic downloadable says the even a one may have just
happened
prize you look back period say its Sam
I think going so im prices a cases
is a tricky one because you will you were prolly hurt you in the future
circa so we have do we talk about performance
a lot of people have again quite very high level
it's got it's a sexy german Finance what's that highest-performing
return that you've ever seen an institutional finance the highest public
1 I've ever heard of his love it that the very things one is
Renaissance technology they're the biggest fund in the US
ultra seeks is highs the best mathematicians and statisticians and
I tight Concord I think they're casket think women are things with the better
I'm they have a huge but
35 percent return every year of fees on every shot okay so that's where absolute
best the best with math PHD's their mind they're they're very
they got a lot on the management process a to them to get back
return with that my money is is incredibly impressive
because you know it's quite easy to get much higher return
if you think a small amount to investigate doing something in the car
high-frequency space
we talked a lot about I quants that leave and they go manage their own money
and their trading sixty your account balances which is a lot of money in real
life
but in in finance it's not a lot of money so a
what kinda returns if they're really good care are possible on those kinda
downturn
with larry i mean you know you hitting the grapevine and people doing twenty
20-25 percent I guess and probably your yet PSAs
time be really wanna be looking at their risk characteristics as well as I suck
figure cannot be created in isolation no of course not but everybody
you know you see it here that use on how much money that I make exactly coming
I would obviously one look at all to see characteristic is a mainland delaney
see how the shop rationing more not but time
I think you you could make a lot of money but the next year you may blow up
busy tight-lipped savior I I think the question really becomes what's the
country's
good return on a sensible the average that is likely lost
long-term and I think you can gently 15 to 20 few really good in exploiting
something that is
actual exploitation %uh the process rather than
so the pub condition right so
with the with quants one thing that caught my attention is you don't have
a magic algorithm like the one algorithm that does everything for you
guy was a sex act upset that you have heart lies and lies about Lindsay got
plenty light is coming out as
the academic would let some stuff that you can begin trading you can
replicate its you tend to have struck you are times where this thing loss to
six months a year
but then you'll know that newbie you be prepared for two dials you have to sort
is
the Act would you even consider it I'm cut off for a trading strategy
night hunting in the woods Hewitt's I'm generally keep it going into
it's not forming disorders level you expect in your
why you would have a mandate you so you basically hoping it doesn't politic off
a cliff that is the case yet generally they do you that the solution slightly
lower ones do tend to just K like that
okay the high-frequency ones can have a bit of a cliff fourth and you can see
why because they're exploiting things like
underlying hardware you right now and if the exchange the size of greatness
hardware or some
some other as incomes and concern seems that so that just immediately destroys
your
your house I heard that's a tough way to what walk into the office and you're
like oh our strategy doesn't work today
hey I think that's why these high previously cuz he pays column ICUs
suffer through the
the the pain that is you know he's the sort six-month research periods where
it just compete in the rugby put up from the room
he's just like it daily do these guys have high-stress jobs
I wouldn't someone I know been in the IPCC's I'd say it's not
much more high stress than any other german Finance the hours on gonna be
crazy
because they need people to be able to think rating right
you know there's a list studies that say that anything beyond a 40-hour workweek
you
come to the ability to missus pretty rapidly and they know that they
allegedly dated June people say they're not gonna make you work $89.00 weeks is
points not look different okay
I'm well not only the investment no it knows that it's all here
and I mean that you know can you be more hours than ever it's certainly been
there with deadlines
clearly but its you know it's much more the
the you know for the performance-driven kind of environment where
was getting things done be making money he we're not gonna make you stay ACLs
with Dalian a height
like yet know is I think about I think there is an A just you know they worked
90 hours a week in a fly around the world it's just that
sounds cool but it's not yet tough game yeah even matter
the so that we'll with a quantitative trait or be up at night thinking about
oh god I got and I i'm only made
percent half this month in my boss is expecting to
I mean its if they're a source smart funds
you gonna have monthly losses are some signs this is the way it is I mean it
and police at a lower frequency yes partly and you will
it's difficult because you know in your practice the really expecting these
periods but as you said you see a nice up with a critique over some
looks great in history but when you actually go through that process if you
to follow the equity curve itself
their periods in its is dropping a land and say
in a wall while you're expecting those things troops it's still painful the
time yes
is what's going on what's happening nothing may be happening because the
statistical properties
alright in school generally civil what you see in the back to assist you
going through the particular period with Israel so is
deeds you look at these tests you get really involved in and it's all
scientific in it has a very
mathematical approach are daequan summer panic
but your too funny I mean it's a you know it's here because he was worried
about what you don't know you have included in the model
this is the guy who are you see here where that you have not created the
perfect model
I think anyone you trusts that model implicitly is
probably not a not really thinking from a fully quantitative mines in another
because it I mean by definition we could be a model is not reality
say this silly things as Miss was a a.m.
and the same here as any proxy so there's always things you don't need any
incoming
if this is the whole ethos behind the Wall Street Crash
2081 in a Salem it
at the models had everything in there the this probably would have been an
issue but
but yeah they did and I mean it there's really famous examples like long-term
Capital Management was one other
the classics they the if they had and it was a plus one hundred
Russian bombs right Russian putting me in that
sexually please him up I section group
yeah exactly but you know that would have been a model Dave expected russia's
the default missus
is quite a Black Swan style event its but he probably wouldn't put it into
practice
so does ok whatever get jumpy
penny I mean it ends what you do you're constantly researching new models in
your you very wears a performance all the time
continue the mission force for you you understand their draw down periods
Unitarians limine is on 07 investors
some cities and those who don't you they will understand if you can get you crazy
drawdown
but OK time it's never fun you know it i think is
she knows they really are lost money a kid except it
concern you know and its you just
you what you really wanna be saying so is the model
doing something was not the right now I'm not aware of is a lot
is whatever's happening summits do the unknown unknown right or is it something
that I'm just
I'm just saying about period and that's the way it might think we're telling me
as I can just come up with an algorithm and then
move to the beach and forget about there's no there's no sipping martinis
early I mean I've had
but this is really funny annotate know the country quite funny needs
two entities function right man in the dog than that in a day
the the man is there to feed the dog and the dog is there to make sure the man
doesn't touch the computer let this tragedy and
is that okay that's what I i mean that's really what I'm asking as I have this
big red button theory
where especially traders that by these robots that I just hit the eject button
at the moment something doesn't feel right
dick wants to the same thing and there's there's a there's a very natural
tendency to want to interfere
okay and is often the case the interfering will generally have a net an
adverse effects
sewing I mean you the classic situation would be
you know you think you models not fully dealing with the antics in the market
opened
quickly okay and say you feel that if you wait say 10 minutes
you might get a you might get better execution that's a that's a
discretionary is awesome senator discretionary overlay and
its is this not hearing to the model I mean you have to say to supplying me is
is my model then
accounting for these enough the the that the dynamics of the market open a
shortened
should I now be putting these into my models getting cool if you sensitive
during that country's micro-management I think it becomes
you know GV is the best as the words whether you're actually running
construction of the next year
I told Mike I think we're at a time yet i want to thank you for joining me yep
and everybody thanks for joining us
I and
I'm gonna cut their yeah done
you
theme