The future effect of banks using AI software | Page 3 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Metatrader using 40% of the CPU all the time when using EA 6 replies

Software to test EAs for MT4 w/o using a demo? 0 replies

Using MAM software question? 0 replies

Anyone using subscription software for trading signals? 1 reply

Anybody using PDA software? 1 reply

Attachments: The future effect of banks using AI software
Exit Attachments

The future effect of banks using AI software

  • #41
  • Jul 17, 2007 12:55pm Jul 17, 2007 12:55pm
Quoting HalifaxCB
I dunno Kudzu, seems to me that one of the joys of FX is that it is really one of the most fixed games in town. I mean, suppose the comptrollers of the big companies got together periodically to talk about what their stock values should be, everyone would have a fit. But get the finance ministers together, it's a major news event.... ditto for CB officials... And wasn't it just last spring at the IMF conference that the heads of the 20 largest hedge funds were called in after for private "discussions"? Can't remember the details, alas. But collusion is the name of the game...
I agree that "the whisper game" behind closed doors goes on now, but still most of the major currencies for the most part are "free floating", otherwise Mr BBB (Big Ben Bernanke) and his economists wouldn't have a job. Also, analyst organizations like University of Michigan and ISM and ADP have a reason for existing because the national economic stats are as accurate as possibly can be expected on a macro level.

However, when countries start manipulating their currencies regardless of economic data and purely on an arbitrary basis then no group of computer algorithms is going to stand up to that.


.
  • #42
  • Jul 17, 2007 12:59pm Jul 17, 2007 12:59pm
Google makes all things possible. Huh?

I will continue to be ignorant and live in the real world and I am not talking the MTV program either.

Technological advances such as phones that take pictures and plays music are a step back not forward. To each their own but the more things change the more they stay the same. Markets included. If, big if, AI rules then there will be no markets. Plain and simple.

Beam me up snotty. Line from Space Balls. Closest I get to scifi.
  • #43
  • Jul 17, 2007 1:00pm Jul 17, 2007 1:00pm
This was the reason I started the thread. Excellent points, WTB.

Quoting WTB
My point is that computers can now play chess better than the world champion, they can fly a 747 better than a 30 year-experience pilot or can play Street Fighter 2 better than an enraged teenager (this last example is a joke hehe). So why would trading be different in a hypothetical future?
  • #44
  • Jul 17, 2007 1:05pm Jul 17, 2007 1:05pm
Quoting tdion
Automated Trading Desk is a firm that trades 6% of the volume of the NYSE and NASDAQ every single day.

They trade strictly on math models with Linux clusters.

They were purchased by Citigroup last week for $640 million.

www.atdesk.com

Word on the street is that they are moving into futures and foreign currencies.
Slick website. How long have they been in business. We're they even around for the last big drop in '00.

I bet LTCM had a nice website too and plenty of PHD's and nobel prize winners.

My point is not to be negative. I'll believe when I see and when someone has done it through ups and downs and for a period of time.

Citigroup is capable of making misjudgements. If you know their history they done it plenty of times before. More smart decisions than not to get to the size they are though.
  • #45
  • Jul 17, 2007 1:15pm Jul 17, 2007 1:15pm
Founded in 1988.
  • #46
  • Jul 17, 2007 1:38pm Jul 17, 2007 1:38pm
Quoting SunTrader
Citigroup is capable of making misjudgements. If you know their history they done it plenty of times before. More smart decisions than not to get to the size they are though.
Very true - in fact one might be tempted to say being bought by Citi is a KoD for good ideas.
Poking through the atd website, it looks mostly to be designed for efficient trade placement. This is really useful for large funds that want to move sizeable chunks of financial instruments that if placed in a single large block trade would move the market adversely - certainly a reasonable thing to do on a computer, but very different than truly AI based trading (and in a sense, much easier).

Like the examples WTB mentions, it's pretty much a bounded problem. Even flying a plane - for all its complexity - is governed by relatively simple rules amenable to programming. Last time I checked - albeit awhile ago - planes were still being flown on chips that were three to four generations behind current technology because of the time lags involved in certification. Even the original space shuttles went up on 8086's. Which gives an indication of the lack of complexity in the problem.

WTB - what you need to do is look at the use of computers on systems of equivalent complexity to trading. This the Big Blue chess computing algorithm is definitely not. A better example would be short term local weather forecasting. I know today they missed again up here - rain predicted, sun received, and in Nova Scotia, that's the equivalent of a big figure move
  • #47
  • Jul 17, 2007 1:53pm Jul 17, 2007 1:53pm
Hal,

interesting that you should bring up the computerized weather forecasting.

Even the current weather forecasting math models that are used are still based on the same old core concept of projecting the future based on the next 10 to 15 minutes. The forecasting of 2 or even 5 days out is simply a long string of 10 to 15 minutes projections piled on top of one another.

Big banks using fancy trading software can run into the same problems that the weather man (or girl) does. The farther you project out into the future the less reliable your computer modeling is.



.
  • #48
  • Jul 17, 2007 2:07pm Jul 17, 2007 2:07pm
Quoting HalifaxCB
A better example would be short term local weather forecasting. I know today they missed again up here - rain predicted, sun received, and in Nova Scotia, that's the equivalent of a big figure move
Who's to say the trades will be "short term?" More often than not, the long term predictions are accurate.

As for ATD, they are trading 6% of the US stock markets every day. That is huge volume. They just raised $60 million in venture capital. To raise that much capital, they must have some evidence that their trading technology works.

I don't believe creating AI for trading forex is difficult. Just look at the EA created for the James16 price action patterns.... those patterns are pretty accurate, and a bank incorporating them into their trading strategies could cause major market changes. For example, can you imagine a 100 billion dollar trade based on a pin bar? I can.

There are other technologies available for accurate forex forecasting as well. A "flexible neural tree" is a second generation neural network that was researched w/ good results on the USD/JPY. You can see the research paper (done by graduate students) by Googling for it.

Whether you deny or accept the implications of AI on forex doesn't really matter. If it happens, it won't be of any consequence to our opinions.
  • #49
  • Jul 17, 2007 2:21pm Jul 17, 2007 2:21pm
Quoting tdion

I don't believe creating AI for trading forex is difficult. Just look at the EA created for the James16 price action patterns.... those patterns are pretty accurate, and a bank incorporating them into their trading strategies could cause major market changes. For example, can you imagine a 100 billion dollar trade based on a pin bar? I can.

EA? Or the indicator? I think A-ha made one(indicator). I only saw early versions but from what I can tell it wasn't exactly very sophisticated regarding placement etc.

Some PA can be automated, some not.

But there are much nicer ways of automation such as ATR based envelopes, moving averages/slopes, etc.
Trust price. Know yourself.
  • #50
  • Jul 17, 2007 2:23pm Jul 17, 2007 2:23pm
nobody expects a truly free market—thats where abuse supposedly occurs the most anyway. on this site there is often talk about compounding on every trade to turn 1000USD into 1000000USD in a year—as we all know, computers are compounding their computing power at pretty phenomenal rates. their ability to aggregate and analyze data is already far beyond that of a dedicated individual. weve said in previous threads on this forum on this subject that short term (subfractions of a second) arbitrage is already almost exclusively in the hands of servers colocated (that is, nearby, asbandwidth latency = lost trades) with trading 'floors'. as with chess, the beginning and the endgames were 'memorized' first.

is a chart with straight vertical and horizontal lines in place of waves in the cards? not if a multitude of strategies are employed by a variety of competing, self optimizing entities.

meanwhile, im still waiting for a strategy to be posted that exploits the workings of a big boys automatic trading system.
  • #51
  • Jul 17, 2007 2:26pm Jul 17, 2007 2:26pm
Quoting HalifaxCB
Tons can go wrong. Good introductions to the problem are in Nassim Taleb's books "Fooled by Randomness" and "The Black Swan"; for wonderful examples of how mathematically perfect systems go wrong, try "A Demon of Our Own Design", by Richard Bookstaber. Both had more or less successful careers in finance, so they aren't just talking theory here.

A major problem is what Rumsfelt (love him or hate him) referred to as the "unknown unknowns" - those events that seem so predictable in hindsight but no one got them right in foresight.

On a more general level, markets and various data exist in a feedback environment; one of the best intro's to this is Soros' "The Alchemy of Finance" (in which he refers to it as "reflexive"). An environment like that can be subject to cascading events triggered by almost imperceptible shifts, which makes robust longer term prediction very dubious.

I think this post needs more attention here. A lot of us are comparing many things to trading (chess, games, flying, etc). However, what many of these things lack is the basic uncertainty exhibited by the financial markets. If human intelligence cannot "out-smart" the uncertainty of markets, how could AI?

Unforseen events subject all players, human and computer, to the same circumstance: unpredictability. While a computer may be more efficient at the actual steps to place and execute orders, the unseen events (unforseen economic news releases for instance) are something that neither a computer nor a human can anticipate.

I think the references to Taleb's two brilliant works (Fooled by Randomness and Black Swan) are certainly appropriate here. I think that almost anyone who has been exposed to trading would more than enjoy them and they would shed some light on this topic.

As far as my understanding is concerned, many AI systems are used in trading to deal with the complex risk models necessary for the large exposure hedge funds have to the markets. Since markets are dynamic, the computations need to be run all the time.

Here's a short example from when I used to work in the futures industry not so long ago. Certain mutual funds offer "commodity funds" which are basically commodity indexes. Since the commodities trade throughout the day, the over/undervaluation of the commodities in the fund need to be constantly evaluted - if corn shoots up in price, for instance, a fund might start selling corn contracts to flatten its position (any one commodity might be restricted to no more than 5% of the fund's assets). Likewise, oil might drop in price and a fund might start buying oil in order to keep the asset weights in alignment. Computers are excellent at keeping a balance between assets in this manner, they also decided when/how much to trade based on the funds' rules. However, the computer was not in the business of "predicting" a rise in corn prices or a drop in oil.

I imagine that even if AI becomes as "smart" as a human (or moreso), it still will suffer from the uncertainty factor all market participants are subjected to.

-=DAVE=-
  • #52
  • Jul 17, 2007 2:36pm Jul 17, 2007 2:36pm
the idea isnt to outsmart or predict the future, it is to make reasoned assumptions and to react more quickly and powerfully than the competition. and believe it, computer models are created for many scenarios.

though i do love how in american disaster movies, always shown are people on the floor of the NYSE in dismay as the market crashes. i suppose that image makes more of an impact on audiences than a computer going 'meep' once and calling it a day. i mean whats next, the jobs flipping soy burgers at mickey deez will be taken by the Earth Simulator? lol


Quoting megadave5000
I think this post needs more attention here. A lot of us are comparing many things to trading (chess, games, flying, etc). However, what many of these things lack is the basic uncertainty exhibited by the financial markets. If human intelligence cannot "out-smart" the uncertainty of markets, how could AI?

Unforseen events subject all players, human and computer, to the same circumstance: unpredictability. While a computer may be more efficient at the actual steps to place and execute orders, the unseen events (unforseen economic news releases for instance) are something that neither a computer nor a human can anticipate.

I think the references to Taleb's two brilliant works (Fooled by Randomness and Black Swan) are certainly appropriate here. I think that almost anyone who has been exposed to trading would more than enjoy them and they would shed some light on this topic.

As far as my understanding is concerned, many AI systems are used in trading to deal with the complex risk models necessary for the large exposure hedge funds have to the markets. Since markets are dynamic, the computations need to be run all the time.

Here's a short example from when I used to work in the futures industry not so long ago. Certain mutual funds offer "commodity funds" which are basically commodity indexes. Since the commodities trade throughout the day, the over/undervaluation of the commodities in the fund need to be constantly evaluted - if corn shoots up in price, for instance, a fund might start selling corn contracts to flatten its position (any one commodity might be restricted to no more than 5% of the fund's assets). Likewise, oil might drop in price and a fund might start buying oil in order to keep the asset weights in alignment. Computers are excellent at keeping a balance between assets in this manner, they also decided when/how much to trade based on the funds' rules. However, the computer was not in the business of "predicting" a rise in corn prices or a drop in oil.

I imagine that even if AI becomes as "smart" as a human (or moreso), it still will suffer from the uncertainty factor all market participants are subjected to.

-=DAVE=-
  • #53
  • Jul 17, 2007 2:57pm Jul 17, 2007 2:57pm
Quoting tdion
Who's to say the trades will be "short term?" More often than not, the long term predictions are accurate.
I'm just using the information from their website. Their specialty is trade placement; its their clients that are deciding the direction. They were formed in response to the market meltdown of '87, which was largely caused by inefficient market placement. A good dissection is, again, in Bookstaber.

Quote
As for ATD, they are trading 6% of the US stock markets every day. That is huge volume. They just raised $60 million in venture capital. To raise that much capital, they must have some evidence that their trading technology works.
Hundreds of millions of dollars were lost in a single day on the portfolio insurance programs that caused the '87 crash. 60 million is, in those terms, small potatoes...
Quote
I don't believe creating AI for trading forex is difficult. Just look at the EA created for the James16 price action patterns.... those patterns are pretty accurate, and a bank incorporating them into their trading strategies could cause major market changes. For example, can you imagine a 100 billion dollar trade based on a pin bar? I can.
It's a very different thing for one trader to put a few K on a pin bar compared to a bank putting, say, 100 million. Your trade (or mine) is not going to have a significant effect on market direction. Not even Nick Leeson could do that, though maybe George Soros could
Small traders - and here I include hedge funds up to say 250 million - tend to ride markets, one way or another. It's analogous to surfing - someone with a board is not really going to change the shape of a wave; someone trying to ride the Titanic onto the beach however will...

Quote
There are other technologies available for accurate forex forecasting as well. A "flexible neural tree" is a second generation neural network that was researched w/ good results on the USD/JPY. You can see the research paper (done by graduate students) by Googling for it.

One should ask why this paper had to be supported by grants from the national Science Foundation of China, rather than from trader's earnings


Quote
Whether you deny or accept the implications of AI on forex doesn't really matter. If it happens, it won't be of any consequence to our opinions.

I think it is worthwhile though (and I think you would agree) to discuss the potentials and pitfalls. From my perspective, trading is littered with corpses of AI based software - and the amusing thing is I am actually playing with something along those lines (using Baysian analysis tools & my Kalman filters). The real problem occurs when people - and here I include such luminaries as the late, unlamented gurus of the previously mentionned LTCM - start believing more in their technology than in their (hopefully) well trained insight and intuition.
Look at it as a simple problem in connectivity - how many neurons can you assign to trading as versus how many nodes a computer can assign? We still win, hands down...
I'm not saying, mind you, that they of no value at all. The big trade block I've already mentionned; another good place is for monitoring unusual pairs for possible misalignment. For a small trader, something like the pin bar EA might be useful, until it becomes common to the level of having a significant effect on the market, in which case it will lose any tradable value. You can see similar effets in other market techniques that were hot for awhile until they caught on (such as the correlation based bond spread trades of the late 90's).
That leaves the algorithm based trader always at a bit of a lag relative to the market - essentially a soft-computing algorithm has to have an already existent history on which to train - and correspondingly it tends to fail to recognize when the trade potential has moved elsewhere. I guess my take on all this would be that since humans are essentially irrational, traders as a group will tend to adapt more quickly (even if individuals don't) than the more complex structure of traders + programmers + mathematicians.....
  • #54
  • Jul 17, 2007 3:11pm Jul 17, 2007 3:11pm
Quoting megadave5000
I imagine that even if AI becomes as "smart" as a human (or moreso), it still will suffer from the uncertainty factor all market participants are subjected to.

-=DAVE=-
I quite agree with you. Basically AI systems have some sort of learning capability embedded in them so that they can learn from past expriences and use them in their future decision makings using nural networks or decision trees, etc.

Now the question is, are these "experiences" limited? In other words can we create an AI system and say after 5 years this system knows how to react to all different market conditions?

We should remember that one the most import driving factors of the markets is the massive amount of emotions and reactions to these emotions comming from huge number of traders all over the globe. What is going to happen if we eliminate this factor through emotionless systems?

Thanks,

Moh
  • #55
  • Oct 30, 2007 9:40pm Oct 30, 2007 9:40pm
Quoting zamans98
Look it from another angle, if computers can outsmart human, they human race are doom, right?

Human create computers, but computer cannot create HUMAN, so that is the TRUTH. The same goes with trading. Forget about FOREX, how about running a small shop? Can computer wins over human? NO WAY...

Look at it from an even better angle - two HUMANS fighting.

One has a spear, the other one has a gun... Who wins?
The one with the guns always wins!

The computer is your WEAPON, how well can you wield it?
  • #56
  • May 24, 2018 3:38pm May 24, 2018 3:38pm
a very powerful thread from a brilliant computer scientist , tdion
i hope smart members of FF will participate & say their opinion

after 11 years of starting this topic , i think we can see many changes & new behavior in forex markets

for example : the member HANOVER has posted this picture in 2009
https://www.forexfactory.com/showthr...44#post2942144

its the 30 days ADR of forex pairs
we see the ADR of EURUSD (for example) was 143 points
nowadays , its just 82 pips for the 30 days average ......what a huge change !!

i hope to hear your oinion
had anyone noticed a change in patterns between 2007 & 2018 ?
was trading easier in 2006 than 2018 ?
Attached Image (click to enlarge)
Click to Enlarge Name: ave_daily_range.gif Size: 20 KB
  • #57
  • Jun 19, 2018 3:29pm Jun 19, 2018 3:29pm
NICE THREAD
  • #58
  • Jun 19, 2018 5:05pm Jun 19, 2018 5:05pm
if computer does the trading, we can only trade with 1:1 leverage, any higher will get us margin call.
  • #59
  • Last Post: Jun 19, 2018 5:33pm Jun 19, 2018 5:33pm
When you think about how far something as simple as a cell phone has come in the last 8-10 years the possibility of a computer outsmarting a trader is very likely. Combine the massive computing capability of a quantum computer with things like voice recognition (for news events), the ability to read and compute an instantaneous response to anything happening in the world, the possibility of computing actual economic numbers along with previous responses. Where computers excel humans are weak. The ability to crunch though terabytes of data in an instant is something a human will never be able to do.

Think about a computer with the capability of listening in on every conversation whether its a news channel or a cell phone doesn't matter and add in the power of the internet with all of it's data. The advances in computing are putting us at a point where even encryption is a futile attempt at keeping data private.

All you have to do is look at the massive leaps in robotic technology that have come about in recent years. Robots exist now that can quickly maneuver through a field of randomly placed objects, jump over obstacles, etc. Just take a look at what Boston Dynamics is doing with robotics.

Inserted Video
0 traders viewing now