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

Why systems work and then don't work...



PureBytes Links

Trading Reference Links

IMHO:
majority of the systems programmed out there are
RANDOM or close to it and what makes you
money is bias or we call it trend plus money management
if any.
entries and exits are mainly random.

what is a random system:
it's a system where the entries are random
no matter what timing technique used.
the parameters used in the system are optimized
to curve fit the bias/trend. this is usually results
in more/easier triggers for buy than
sell signals in up trend and vv. in down trend.
easier means conditions for buy are loosened
relative to the sells.

this is proven by observation of the following
( applies to discretionary and systematic trading ):
- markets are mostly unpredictable
- markets develop trends or biases.
- most entries are random
- most exits are money management
the most common trading approach is this:
- find common trend or bias
- establish your positions by trading with the trend.
  buy in up trend and sell in down trend.
- wait for two things to occur
  1. position goes profitable - take profits
  2.  trend reverses - take losses.
trend is present will make you money and your
simple money managment.
majority of people trade this way knowingly or not
( if you are rational trader )

i am going prove it to you that it's the way it is.
below is a simple random system where you
can bias it against or in favor of the trend.

{ $RandomMoney }
input: bias ( .5 );
vars: trigger ( 0 );
trigger = random(1);
if trigger < bias then
 buy;
if trigger > bias then
 sell;

this system has one input that controls how biased
the system is towards buying or selling:
the input called bias, the range of that input is
from 0 to 1. the default value is .5 that means
system is unbiased towards buying or selling.
the system entries and exits are randomly generated
( normally distributed ) value1 = random(1);
random function generates a value from 0 to 1 on
every bar. then by tilting bias input of the system
we can bias the system to easier buying than selling
if bias > .5 it will be easier for the system to buy
rather than sell. if bias < .5 it's easier to sell.
we are trying to optimize this by biasing the system
to or against the price trend...
ok. so far.

now let's take the bull market in nasdaq for about one year
1/1/1999 - 3/10/2000 daily nasdaq composite.
and apply this system to it.

case1: bias = .5 (random entries and exits)  no bias.
do we make money?  sure we do in most runs... we got about
50% profitability ( proves our system is unbiased )
we make money in most runs!!! why - the trend, the trend....
our buys are simply more profitable then sells
because we are moving up more than down ( up trend ).

case2: lets bias ( optimize it a bit ) it to easier buying:
bias = .9 do we make money? we sure do...
do we make more money?  we sure do in most runs.
( one note! since system is random - system results are
random from run to run )

case3: lets bias it against the trend towards selling easier
then buying.
bias = .1
do we make money? nope.. not anymore in most runs.
how come? well simply speaking we are going
against the trend more times that not...

that's not it yet. ok. like smart systematic traders we
are before 3/10/2000, lets optimize our "great system"
by simply biasing it to max profit by selecting our
bias input to .9, right? ( that's where we make most money we think )
the system is pretty good: it makes money.
heck, let's keep this parameter .... :-)
(we are gonna go to  the moon, dow 20000 nasdaq 10000, )

here comes what comes after 3/10/00
set your chart to start from 3/10/00 till now
do we make money with the same .9 parameter?
hell no in most runs , system turns to crap.
because the trend changes from up to down.

conclusions:
1.trend is what makes you money... not your parameters
or entries or exits or timing. it's easy to set a trap by
optimizing your parameters to fit the trend.
they sure help but they are not as important as bias or
trend component. no matter how complex your entries/exits
the chances are they are going to be random ( 50/50 chance )
2. design your system to recognize trend and then you
can adjust  your parameters accordingly..
3. the systems stop working because they were fitted,
biased or optimized based on the prevailing trend...
when trend changes, system stops working if parameters
are biased
 if parameters are unbiased the system will make money
in any trend up or down but not in sideways trend.
in sideway it will breakeven ( random in random ).

case4. lets now change the parameter back to .5
it same chances of buying and selling.
do we make money? in most cases we do...

so what's the best parameter - 0.5? nope.
because if trend turns to sideways we stop making
money- we will likely to break even.
the best parameter:
trend up - .9
trend down - .1
trend sideways  -.5
so parameter = f ( trend ); ......

real example:
one of my acquaintances coded and traded the stock system before
trend changed in nasdaq in march. he thought he won lotto -
twaz so easy to make money. system called the shots like magic.
after trend changed - big losses...
he now abandoned the system completely. why because
if there is trend!!!! any entry, any timing technique that
had rational  worked! ( buys of course in up trend ), entries simply did not
matter. exits were mostly profit taking ( a form of money management ).
you get high and think you got the best system there is. you could
throw darts at it and win huge. free money - it was!!! not anymore.
as soon as trend changed - the parameters
stopped reflecting the trend - huge  losses came in because the
system kept buying in down trend...

most systems out there are random by nature.

characteristics of a "random" system:
- makes money in trend / loses money in counter trend with same
  parameters. ( curve fitted ) OR makes money in any trend but
 has drawdown and around 50% profitable ( over simlified / generalizing
system )
- complex or oversimplified entries / exits ( random irrational )
- poor management ( no money managment or too many parameters )


characteristics of a good "non random" system:
- makes money in trend, counter trend or sideways ( parameters
are adjusted based on the trend ), identifies trend, more than
50% winners in any trend.
- simple but meaningful entries and exits ( biased with trend )
- money management based on trend recognition ( when trend
reverses it's time to exit the losing position else  take profit in
direction
of trend ).

bilo.
argue or not - just load in the system set the parameter, do some runs
and look for yourself.
because this system is random the stats will be different on every
system run, but the biases will still be reflected in the stats over several
runs.
the best way to do repeat runs is from power editor by clicking "verify".
current trend is up. or is it?:-)
have a nice 4th.