Random Entry & Perception

[DionysusToast]

I guess the bottom line with what I am trying to say about their test.

It will work in trending markets much like an MA crossover system. If I took 10 different markets/stocks and showed you an MA crossover system with a trailing stop and showed it worked on all 10 over the timeframe selected and then said "MA Crossover systems work", it would be correct.

Now - I am sure most would come back with an argument about handing all the money back in choppy markets and they would be correct. As would people that said the test was across too few markets and too little time. Then again, we all know about MA crossovers. The thing is I never said MA Crossover Systems work in all markets all the time. In fact, the readers may infer this based on the way I articulated the results but still I didn't say it would work all the time.

Similarly in the Van Tharp model, they never said it would work all the time or in all markets - but they made such a song and dance about when it would work - it is reasonable for someone to jump to conclusions. It is possible they jumped to conclusions themselves.

The system is a trend following system. It relies on trends lasting long enough for the bad coin tosses to eventually be followed by a good one and to grab a large bite of the trend. In the 20% of the times where it didn't work, they were hit by a bit of bad luck where the coin tosses didn't go their way. They suffered from a few too many bad coin tosses and the good coin tosses couldn't capture enough of the trend to catch up.

So - ultimately the test doesn't prove what is implied but then they don't absolutely state it will work in all markets as we may presume ourselves.

 

[DionysusToast]

Meanreversion

If we say 95% of people lose money but a coin toss would make money 50% of the time, we have to consider a few things.

The coin toss will give you 50% winners ONLY if the distance from entry to stop is the same as entry to target. Slippage and fees would take the break-even approach to a loss. Which means 100% would lose money following this approach. They would not lose money 100% of the time but 100% would be losers.

So in essence - even the most dedicated to$$ers such as 'Insert name here' will lose money.

It does have to be said though that these losses need not be distributed evenly. Over a large enough sample size, we could reasonably expect one to$$er to be right 5% of the time but another to$$er to be right 95% of the time.

This distribution points out another obvious flaw in the test. After setting up the initial rules, they suffered from losses 20% of the time. This is because the system couldn't handle the randomness or rather it soulcn't handle uneven distribution. It needed fairly even distribition of heads/tails. When they didn't get that, the system failed and they ended up tweaking it to fit the data.

 

[Aussie_Pride]

Happy to be ridiculed so here goes:
The auther may be trying to make "holistic" points about trading strategies (for example KISS) but I do not believe they are pertinent to this discussion.

I have to agree here. While it maybe something that is done, i think the main point was on keeping it simple, that we don't need to complicate things to make money. We don't need 20 indicators on a chart and all the rest of it.

My two quick cents

 

[BSD]

Meanreversion ran NEW backtests to clarify the point to himself, hence he'd hardly be fiddling the data lol !

Same like this guy who ran NEW tests that were profitable 100% of the time:

"As Anthony suggested I run 100 tests of the random entries-trailing stops at 10 ATR from 1982 until September 2007 on a portfolio of 69 futures with $100 for commission and slippage and the system made money 100% of the time as you can see in the following figure."

http://www.tradingblox.com/forum/viewtopic.php?t=3637&postdays=0&postorder=asc&start=0

This is really all it's about, hard, backtested, objective facts.

Either one is in complete, total and utter denial, or one isn't.

And of course trend following systems only work when markets are trending, what an amazing insight.

This has got to be the most pointless thread I have ever seen on T2W.

Who you guys gonna believe, some benchmarks who are at the top of our field, a buncha Market Wizards who got rich from trading, and nothing succeeds like success does it, and who really have all the resources needed at their disposal to research and backtest...

Some guys who trade for a living and did independent backtests like our Meanreversion or the guy above...

OR or some really sharp guys like Gecko who fell for wasp and had him trade his money for him, and who only ever shows up here with some cryptic posts or statistical mumbo jumbo...

Or a really clever guy like Toast constantly regurgitating his obscure theories who believes Pros yo don't use charts, or, just as delusional, goes round coming up with closed-minded general stroke conclusions about TA or indicators just because he can't get them to work for himself:

Regular/Regurgitated Technical Analysis

You know MACD divergence, MA xovers, Stochastics - all of those common entry techniques often discussed in books on trading despite the fact they don't work.

I think this covers anything you read or learn in 99% of books & courses.

And yet is himself straight out of some course from some guy here at T2W.



Besides as I have shown numerous times also in this thread both of those assertions are completely ridiculous anyway.

Guy who wanted to land a job at a bank called Brett Hipkiss got a degree, opened an account trading currencies, doubled that over the next couple of months, on said track record got an invitation from HSBC, and in the interview explained how he'd done it using technicals and indicators, and that landed him a trading job there. (Trader Daily, May/June 2008, p.52)

Or Maggellan here with his 50 000% in one month who used and uses charts lol...

Same goes for this hedge fund running 1,5 billion.

No contest is it who you wanna believe.

If I wanna be a great golfer I'd go study Tiger and cohorts, and not listen to some posers spreading either statistical mumbo jumbo and obtuse, unproven theories instead of backtests and completely made up untruths on message boards.

If anybody here thinks meaningful conversations are possible with people who, in the face of all evidence to the contrary, believe the earth is flat I certainly do not belong to them.

 

[Black Swan]

Well - I shall tell you why.

It seems we have a tendency to believe the written word. If I'd read this book and not merely been pointed towards this pattern, then I'd have probably spent a few minutes reading this passage and then moved on. I'd have more than likely believed what it said and then, taken it for gospel and maybe even repeated it as fact.

The problem is, it is flawed.

First - let's consider what this study ISN'T... It is not a Blind Experiment...



The bottom line is that an expectation of a particular result leads to bias in testing.

Interestingly, another psychological oddity is the fact that I could not write the first post in this thread without alerting you to the fact that something was wrong. It would be impossible for me to bring this passage to your attention without also bringing to your attention the fact that something was amiss (or I wouldn't be asking the question).

As it is, there are 2 major flaws in this study :

1 - The approach taken to the study is flawed
2 - The test itself is NOT proof of random entry, rather they admit to curve fitting as well as omitting key data

Given the text - does anyone else see why the approach is flawed and why the entry is in no way evidential of randomness + MM = profit ?

Also given the text - does anyone have an opinion on whether the authors intended to deceive or whether they deceived themselves in the process. The flaws we can prove, this we can't but I would be interested to hear your thoughts.

Good points...

 

[Black Swan]

Meanreversion

If we say 95% of people lose money but a coin toss would make money 50% of the time, we have to consider a few things.

The coin toss will give you 50% winners ONLY if the distance from entry to stop is the same as entry to target. Slippage and fees would take the break-even approach to a loss. Which means 100% would lose money following this approach. They would not lose money 100% of the time but 100% would be losers.

So in essence - even the most dedicated to$$ers such as 'Insert name here' will lose money.

It does have to be said though that these losses need not be distributed evenly. Over a large enough sample size, we could reasonably expect one to$$er to be right 5% of the time but another to$$er to be right 95% of the time.

This distribution points out another obvious flaw in the test. After setting up the initial rules, they suffered from losses 20% of the time. This is because the system couldn't handle the randomness or rather it soulcn't handle uneven distribution. It needed fairly even distribition of heads/tails. When they didn't get that, the system failed and they ended up tweaking it to fit the data.

I'm tempted to lower the tone of the thread by saying something such as; only a teenager could regularly toss themselves of that much...but...oh well...

 

[BSD]

Oh brother don't encourage him, I'm really starting to think I'm in the wrong movie / board here...

WHERE can Toast prove his assertion that:

"2 - The test itself is NOT proof of random entry, rather they admit to curve fitting as well as omitting key data"

I guarantee you anything that is a straight out lie from toast again.

I really will try and contact Tom Basso if Toast upholds that lie.

The ONLY thing Market Wizard Tom Basso wanted to figure out was whether random systems can work beacuse some one had posed that question to him...

WHERE ON EARTH can one fit some conspiracy theory into that like Toast with his fact twisting is TOTALLY beyond me...

Besides, Toasts BS completely ignores the NEW backtests of Meanreversion and the guy I quoted above from Curtis Faiths board that verified Bassos findings.

 

[DionysusToast]

getting an 80% success rate and then adding new rules to bring the success rate to 100% for the same data set IS curve fitting.

As for the omitted data - the SIZE of the losing 20% to the SIZE of the losing 80% is omitted. This is very pertinent data. Also - the SIZE of the winners after adding in the new rule is pertinent data.

Any change you make on a backtest to increase the win ratio without then testing on another dataset is curve fitting.

You can contact Basso, Van aTharp and Van Helsing if you like, it does not change the fact that they broke some fairly fundamental rules of system development. Rules that you may not be aware of BSD but very important rules nonetheless.

 

[BSD]

Again: YOU made the ludicrous and preposterous claim that they admit to curve fitting as well as omitting key data.

Back that up.

I GUARANTEE you are lying.

Why ON EARTH would Basso fiddle the data when he just wanted to clarify a point, can random entry end up net profitable or not, that is ALL he wanted to figure out.

Trying to invent a loony conspiracy theory about that is just unbelievable !

And for the UMPTEENTH time...

You are completely ignoring that Meanreversion and the guy from Curtis Faith's board CORROBORATED Bassos findings with NEW tests across DIFFERENT markets, 69 in the case of Curtis guy.

 

[BSD]

Come on already, 10 minutes and still no link from you, you make a ludicrous claim, so provide proof in form of a statement you can link to where Basso admits that he curve fit his test and omitted key data just cause he wanted to hoodwink some people so you could call him out years later.

You make a preposterous, slanderous claim like that, so BACK IT UP !

Unbelievable how some one can be so delusional as to try to build a case on totally flawed premises...

A guy who got rich from trading - Basso -runs some backtests to clarify a point for himself, and is then supposed to lie to himself by twisting the data ???

 

[DionysusToast]

BSD - it is not a conspiracy. You have comprehension problems. As I have mentioned before, you appear to read what you want to read and are not able to understand nuance.

For instance, when I say "I know a day trader that doesn't use charts", you hear "DT says all pro traders don't use charts".

It is pointless you coming here and debating because you are not arguing against me, you are arguing against what you think I said. There seems to be a problem with your perception of my written word because you already have some pre-conceived ideas about what I am going to say.

If I said "The sky is red tonight", you might reply "The sky is Not purple, don't be ridiculous", then you'd post some downloaded pictures of a blue sky, a picture of a nice car under a blue sky, some blue text and a youtube video of the sky.

Like I say - there's no conspiracy. It is merely a common mistake. People curve fit all the time without realising it. People develop systems that look fantastic on a backtest that fail in reality. I did not say that anyone intentionally curve fitted.

What they did to get the extra 20% is absolutely without any doubt, curve fitting.

Now - please post the methods of the other tests to prove random entry works and we can pull those apart too.

 

[BSD]

Again: YOU made the ludicrous and preposterous claim that they admit to curve fitting as well as omitting key data.

Back that up.

I GUARANTEE you are lying.

Why ON EARTH would Basso fiddle the data when he just wanted to clarify a point, can random entry end up net profitable or not, that is ALL he wanted to figure out.

Trying to invent a loony conspiracy theory about that is just unbelievable !

And for the UMPTEENTH time...

You are completely ignoring that Meanreversion and the guy from Curtis Faith's board CORROBORATED Bassos findings with NEW tests across DIFFERENT markets, 69 in the case of Curtis guy.

I don't want to hear any of your obscure theories or beliefs that the earth is flat.

I don't deal in fiction, I deal in facts.

You made a slanderous claim about Basso, and now I want to see proof where Basso admits according to you to curve fitting and omitting data just cause he likes lying to himself !

 

[MrGecko]

If I said "The sky is red tonight", you might reply "The sky is Not purple, don't be ridiculous", then you'd post some downloaded pictures of a blue sky, a picture of a nice car under a blue sky, some blue text and a youtube video of the sky.

Sorry but... LOL!

 

[BSD]

Yes lol I'm sorry too that that sort of disingenuous fact twisting smoke and mirrors while always avoiding anything concrete in the form of value adding hard facts is really all Toast or colleagues like Socrates did here at T2W.

This entire thread is predicated on Toasts claim that he is privy to Bassos test methods, and that Basso admitted to curve fitting and omitting key data.

That, folks, is a blatant lie, and is why Toast is not able to back up that slanderous claim.

How preposterous to assume that Basso would lie to himself when he was just on a fact finding mission that ended up in the public realm not because of himself, but because Van Tharp published the findings.

Toast, cut the disingenuous dissembling and BACK UP YOUR SLANDEROUS ALLEGATION WITH A LINK !

 

[meanreversion]

Disingenuous - one of my favourite words. BSD, your tirades are wonderfully eloquent, I salute you!

Guys --- all take a chill pill. What was the original point of this thread - random systems can't work, was that it?

Ok, take a step back. Define "random system". It's almost impossible to conceive of a totally random system.. this would be one which randomly entered in either direction, in a random amount, on a random market, and then at some random point in the future, randomly exited a random proportion of the original trade.

The claim is NOT that such a completely random system would work.... in fact, it's not even worth testing as no-one would ever trade in that fashion (e.g. ask your dog whether to go long USD/JPY, if it barks once, buy 1 unit, if it barks twice, 2 units, if it whimpers, go short 1 unit etc etc). This is clearly a nonsense and is a) not what Basso claimed would work and b) is of no interest to anyone, so let's stop talking about it.

The random ENTRY system has a very defined and specified set of rules on position sizing and exits. The exit will be done as a trailing stop, based either on a Donchian channel or a "Chandelier" stop (based on x ATR away from n day recent high).

This PARTIALLY random system will make money over time if applied to sufficient markets, due to the existence of trends out there somewhere.

The conclusion is that sizing and exits will make or break a system, despite an almost universal lack of interest in the two. When do you hear on CNBC "now's a good time to sell your Goldman stock"? Maybe once in a while, but the other 95 pct of the time it's "buy this, buy that".

Simmer down everyone!

 

[MrGecko]

A tangental point, but I reckon it can be shown that a fully automated random entry strategy will not outperform the market over time.

 

[meanreversion]

Ok, I ran some tests. The "proper" system enters on a break of the 20 day high/low and scales in up to a max of 2 pyramids as the trade goes further onside. A moving average trend filter determines whether to trade or not. The exit is if the 20 day high/low is breached (for a short/long position).

I used 9 markets (mixture of FX and commodities), maximum of 7 positions at any one time, and ran the tests over the last 10 years. Amibroker is the software.

The "random" system decides to enter on a coin flip (heads it trades, tails it doesn't) but the sizing and exits are identical to the proper system. For one run I applied a trend filter to the random system, on another run I didn't.

The "proper" system returned a CAGR of 37.19 pct per annum. The results I got for the random system were as follows -

1. With trend filter
34.46, 40.86, 38.21, 27.88, 27.78, 39.23, 30.63, 44.30, 41.42, 29.38; average = 35.42

2. Without trend filter
21.89, 21.53, 14.39, 31.62, 15.99, 20.89, 12.26, 20.86, 13.48, 12.19; average = 18.51

Two things of note here

a) All runs for the random system are positive, irrespective of filter
b) The random system with filter sometimes outperformed the non-random system, BUT on average is less profitable (one would hope this is the case).

 

[Hilarymannah]

Here's an example of something most folk don't see,..but it's there, regardless of how many opinions are discussed. The arrow in the FedEx logo:
http://sportskate.files.wordpress.com/2009/03/fedex-logo.jpeg

It's hidden in plain sight ! : )
BTW: I've loved reading your debates, Toast and BSD,..thanks for the entertainment

 

[Shakone]

DionysusToast: As I have mentioned before, you appear to read what you want to read and are not able to understand nuance.

That might be true, but I have to say that you've made an awful job of getting your point across.

First you mentioned random system, which meanreversion pointed out doesn't make any sense, and not even relevant since we're talking about random entry only.

Then you make ludicrous statements like

DionysusToast: There is a reason that random strategies don't make money - it's simply because they can't. When I say they can't - I mean it is against the laws of mathematics and in a way physics.

My random entry system is to enter once only, and take profit at 10 pips, and my stop at 10 pips. CAN it make money? Of course. You are picking the wrong words to explain what you are trying to explain. There are no laws of mathematics that say you can't make money with a random entry, but there are laws of probability which say that you CAN. And if you can win money on one trade, you can win on 10 or 20 or 30 in a row.

So I can only guess that what you are actually saying is
1) The experiment in the text was flawed
2) That a random entry system will not make money in the long term (i.e. over a high number of trades) in all markets and in all conditions.

Well for 1 - - Agreed, it isn't even clearly explained

And for 2 - not every successful system makes money in all markets under all conditions. In fact, I'd guess that most successful systems don't in all markets - although I can't prove this.This is not really the fault of random entry. But you certainly haven't proven that with the right exit criteria and money management that it CAN'T be done. The reason you can't prove that, is because it is not provable. And as someone who claims to understand maths and physics, you should realise that there is simply no way you can prove that over ALL possible exit criteria and money management rules, a random entry system can't make money in the long term.

You would have to prove for EVERY (and there are infinite) exit criterion, you can't make money in the long term. Whereas those who say you can make money, only need one counter example, and you've already been given plenty.

If you are arguing something else, then try to explain exactly what you are arguing, otherwise you get a thread where people are arguing different loosely related topics, and as a result not agreeing.

 

[MrGecko]

not many systems make money in all markets under all conditions. This is not really the fault of random entry. But you certainly haven't proven that with the right exit criteria and money management that it CAN'T be done. The reason you can't prove that, is because it is not provable. And as someone who claims to understand maths and physics, you should realise that there is simply no way you can prove that over ALL possible exit criteria and money management rules, a random entry system can't make money in the long term.

It is beyond the reaches of this thread, thats for sure, but if I were someone fluent in mathematics (alas I am not) I would attempt to address this problem in two steps:

i. Use a "no arbitrage" argument to show that, in the limit, entry and exit criteria are all equally profitable.

ii. Given (i), I would then integrate over the whole time series for a particular strategy.

This, I believe, would yield no more or less than the risk free rate or Beta returns.