[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

counterargument to c.lebeau's constant bet size under drawdown--true market risk management



PureBytes Links

Trading Reference Links

dear chuck,

the first argument for urging care under a drawdown is because of the
nonstationarity of the returns distribution curve of a system, due to the
nonstationarity of the underlying market.  gambling examples don't cut it,
because a game is inherently stationary--and this ain't the markets that we
trade.  this characteristic of markets was talked about, for ex., in the
cult classic "the predictors".  the result of this characteristic is that
the majority of models blow up in actual trading {or even forward testing}.
this is a big part of why this is the common experience of old system writer
plus and ts users.  the book details how these quants got so desperate
trying to find models that worked even in testing that in crunch time to
produce proof-of-concept to their investors, programmers ended up "throwing
everything at the wall they could think of to see what would stick".  this
all sounds too real and too painful.  the reason to wind down bet size under
drawdown is to reduce the potential adverse impact of the nonstationarity
working against the trader--i.e., the famous out-of-left-field negative fat
tail that comes to bite the trader.  this is risk management.

this risk can be greater for a trader that trades several markets.  the
classic example is being in 4 or 8 markets say all on the long side. then
the market reverses and all our positions sour at once.  this intermarket
correlation--which admittedly varies from day-to-day--seems like murphy's
law of trading to come all at once.  utilizing models based solely on
testing one market ignores these intermarket risks.  now the traditional
answer is to mix one's portfolio say half on the long side and half on the
short.  again, this rule of thumb fails to examine inverse correlation of
markets.  one could easily construct such a portfolio that would end up in
worse shape than being just in one market.  again, intermarket correlation
with multiple positions under drawdown suggests smaller bet size more in
line with ones money management system.

further, what any trader that have been around for a while knows is that
markets are constantly evolving and changing.  new trading strategies
diffuse as more and more participants learn them.  electronics speed up info
diffusion as well as trade execution.  globalization increases
interdependency of markets.  etc.  what much systems modelling fails to do
is to make explicit what the assumptions of markets are under which the
models that are developed operate successfully.  so if and when markets
change adversely in some structural way, one can decide whether to throw out
a model despite its working well [or more likely, when the model craps out,
the trader realizes that the markets have changed].

i still feel that this is a critical area of neglected research.  mechanical
system users all too frequently simply say that if my system works, keep
using it.  if it ain't broke don't fix it.  if it stops working, find a new
system that works.  this is pragmatic, but not necessarily enlightened.  it
seems to me that a much more productive approach would be to ask why does
this system work?  what does this system assume about the market and how it
operates?  what are the falsifiable tests of the market assumptions?  what
are the defining (and quantifiable where possible) characteristics of a
market that can help aid us to categorize markets?--in a manner that will
help us in the selection decision about which model works in this type of
market.  we have crude ways of describing this now:  this is a trending
market, this is a sideways, countertrending market, this is a hi/medium/low
volatility market...this is a practical step in the right direction, but it
seems to me much more can be done here that could significantly enhance the
effectiveness of our systems trading.  our understanding here is woeful.
how many mechanical system traders know the market assumptions of the model
that they are trading?  rather than trying to find markets for which our
models work by trying them out, maybe we can use our pc, parameters and
market modeling to tell us before we have to get burned--or miss out on a
market we might not normally think of.  let's be proactive to
nonstationarity, not reactive.  learn not just the system and just the
market, but their interdependence, too.

now regarding market meltdowns.  lately, i've been doing some reading in
garch models and their applications.  [see the omega spinoff group,
behavioral finance, for ex.]  without going into a lengthy discussion here,
as you suggested, strictly speaking even garch models, which are
conceptually correct and far superior to most conventional models which are
based on emh theory, are still poor in predicting market meltdowns.  they
operate over a larger range of distributions.  they have a better estimate
of market risk.  they are great in after-the-fact estimation of true market
risk.  but they are poor predictors.

assuming that market meltdowns are random, however, seems to me to be
self-defeating.  we may not be able to predict exactly when they happen, but
what we can do is to identify environments under which our model will blow
up.  in these environments, our market risk jumps up.  if these materialize,
then lighten up the bet size.  so if the market dramatically turns, then the
trader will not be hurt that much.  or if the threat is catastrophic, get
out completely and just watch from the sidelines.  or--if you are really
sophisticated--get ready with whatever alternative system you might have to
take advantage of the structural change.  this, imho, is real risk
management--the type that matters over the long run.

what is useful as the garch model research has affirmed is some variation of
worst case market scenarios.  that is, use some thinkits [finance theory
version of einsteinian thought experiments].  these 'market breakdown
hypothetical situations' have at least two basic categories.  first,
consider the dominant paradigm(s) underlying the current market.  isolate
its key assumptions.  quantify if possible.  figure out at what levels these
key assumptions going forward would be violated.  consider some critical
mass of violations that would falsify the prevailing paradigm.  then this
marks a key structural break in the market and a potential turning point and
potentially dramatic reversal and model failures.  the new economy paradigm
driving the stock market for the last decade is a good example.

second, create scenarios in which all hell breaks loose.  this category
includes what may be considered exogenous variables: institutional impacts,
politics, war, etc.  these are things that may not now be in effect, but if
they come to pass they can dramatically affect the market that we are
trading.  should a thinkit materialize, what would you do to manage the
risk?  who had any prior experience with the gulf war?--tho everyone knew
what would happen to oil prices.  because of the weeks for deploying intl
forces we had lots of lead time.  who had any prior experience with the
asian financial crisis?--tho even the man-in-the-street knew that whatever
it was it was serious.  get out of stocks.  isn't this what the drums were
beating--for mos?  what would happen if the fed changed margin rates on
stock buying?  what would happen if a nuclear bomb went off wherever?  etc.

now in the second category, sometimes we get some notice--often a long
notice lead time, only nobody's listening.  well, you listen.  then we can
choose to lighten up, get out or watch for further confirmation.  sometimes
we don't get notice--then the premium is on reacting fast.  and as any
trader knows it is much easier to pull the trigger with a trading plan than
it is to reactively try to deal with ones initial shock and disorientation
and then to figure out what the hell to do as we watch paralyzed as our
positions sink.  if you think this only happens to beginners, just imagine
druckenmiller watching verisign tank--even loading up on his position big
time--in the middle of the nasdaq correction that started in late mar in apr
but well after the net space was being bashed from at least the summer
before.

again, if you are in a drawdown during one of these thinkits starting to
unfold as reality and that would blow up your capital accounts--at least
lower the bet size.  better still, get out FAST!  the worst you can lose by
going to sidelines is some profit if you are wrong and markets resume its
normal behavior.  the worst you can lose by staying in and ignoring the
market risk signs is a wipeout.

so on the contrary, sometimes we are lucky, the markets actually give
signals to the trader and time to exit or lighten up prior to a meltdown.
sometimes we will get stuck with a somewhat larger position without
warning--but if we are following a sound money management system [meaning
can we take the hit if all our positions take a bath?], then we can take the
loss.  [unlike the go-for-the-moon manics, i do not usually believe in
betting a lot on one bet. that is a strategy to be used only for the
once-in-a-lifetime bet.]  more to the point with these thinkits, we can exit
faster and loose less.

even Long Term Credit had 4-6 weeks lead, but made the wrong decisions.
they ignored the counterparty risk of their derivative arbitrage positions
and the impact on liquidity of a market meltdown.  instead of lightening up
on their positions by exiting more of their high-risk liquidity positions
and drastically increasing cash, they held steady.  but that's another
story....

in summary, it seems to me that risk management fundamentally means to take
fewer risks when the market risk environment worsens or can dramatically
worsen shortly--particularly in drawdowns because that may be the beginning
of a dramatic shift against your favor.  we cannot forecast the market.  we
can only manage the risk that we take.

regards,
brad yoneoka


> -----Original Message-----
> From: CRLeBeau@xxxxxxx [mailto:CRLeBeau@xxxxxxx]
> Sent: Monday, June 12, 2000 11:47 AM
> To: omega-list@xxxxxxxxxx
> Subject: Re: counterargument to c.lebeau's constant bet size under
> drawdown
>
>
> n a message dated 6/12/00 10:49:09 AM Pacific Daylight Time,
> byoneoka@xxxxxxxxxxxxx writes:
>
> << so these considerations: we can't/shouldn't try to forecast the market
>  {implicit in choosing to trade a given system}, we never know
> when a market
>  meltdown will occur, and we can choose the priority of trading to manage
>  risk--all argue against constant bet size under drawdown and for
> asymmetric
>  money management.
>
>  regards,
>  brad yoneoka >>
>
>
> Brad,  I agree with most of what you said.  A very informative
> post.  (I'm
> not sure your post made it to the list so I am pasting the full
> text below my
> reply.)
>
> However, what happens if the market meltdown you refer to occurs when you
> have substantially increased your bet size?  Isn't that really
> the worst case
> scenario?
>
> Seems to me that we both agree that we don't know when the meltdown will
> occur.  But isn't this in fact an argument for a constant bet
> size that will
> not have us risking too much just before the meltdown?
>
> Chuck LeBeau
> traderclub.com
>
>