"TheInternet" Bot - Doing Something with Trades - How long? What happens after?

[BitcoinPorn]

Posted this


Replied somewhere was this

The bot has been off line the last couple of days for upgrades and debugging. I've already made the trade system more robust and will be adding some market detection code today so the system can automatically switch between radically different GA strategies on the fly.
That being said, it has already completed about 600 trades and is profitable

http://www.reddit.com/r/Bitcoin/comments/ipr35/bots_keeping_the_price_still/c25oud3

Then I had to go look and found this

and this

I'm a thinker, an architect (INTP), not the business type.
I'm not doing this to make money -- neither through selling the software or making money on the market, I'm only doing it because I find it interesting.
At some point I will probably get bored and abandon the whole thing.

http://www.reddit.com/r/Bitcoin/comments/ijp4s/high_frequency_trading_with_genetic_algorithms/c24bi4v


Awesome right?

At the least, better than this http://forum.bitcoin.org/index.php?topic=28268.0 ?

I know this shit goes on, just have not seen the screens before is all.   I could possibly title this post better, but leaving for now, still reading on how this particular bot works.

[1713925001]

1713925001

[1713925001]

1713925001

[1713925001]

1713925001

[1713925001]

1713925001

[1713925001]

1713925001

[1713925001]

1713925001

[TraderTimm]

Love the 'net worth' part of the graph. Mimics price, which means it isn't any more efficient than price. That doesn't hit me as a particularly effective system. Most system makers want their equity graph to have a nice upward trend, not full of drawdown excursions when price falls.

I give this a big 'meh'.

[Big Time Coin]

Thanks for doing this research.  Don't be quick to condemn trading bots.  They are inevitable and ubiquitous in all mature electronic trading markets.  They also provide a valuable service to human traders. Rapid filling of orders, liquidity, and executing algorithms to determine fair market value.

There are drawbacks like flash crashes.  But with a low volume, low market cap market like Bitcoins, it seems unwise to make a large investment or to attempt to redeem a large investment (let's say $10 million and up) without the help of an automated trading platform to spread out the purchase or sale over many weeks so the market can bear it.

[HideousBeastManGuy]

You guys really overestimate the impact of bots on bitcoin, it's minimal, and most of the impact is positive, not negative.

I read through that entire thread on reddit yesterday.  It's a Genetic Algorithm.  Basically you create a pool of "genes", which are in this case, trading methods, and you let them exist in different populations under different conditions, with given starting parameters.  From there you see which are the strongest (biggest winners) and you let them merge with other genes to hopefully create a powerful winning strategy.

The guy really doesn't give a fuck about profit, and I actually believe him when he says it, he's just a dude with an interest in cutting edge algorithms.  His sample size was 5 trades as of yesterday, and his winrate was 80%.  This isn't definitive at all.

Also, I'm not sure I subscribe to the whole genetic algorithm being something that can be applied to markets, as it seems to me you'd have a massive hurdle in that the algorithm would want to choose the best winners on a curve-fitting basis.  Perhaps he's taken measures to avoid this, but the main point is that what worked yesterday, has to work today, in order for an algorithm based on past experience to work.  Clearly the fact that there are winning traders who trade purely Technical Analysis suggests that this exists, but they definitely don't curve fit.

EDIT: Oh yeah, it threw me way off that he called it "High Frequency Trading" because it's not even close.  In fact, this was the first time I noticed.  You cannot run HFTs unless you have Fill-Or-Kill (FOK) orders.  The only exchange that has that is CampBX, and market depth is shit right now on that exchange.  An HFT probes for information on incredibly short term supply and demand, and gobbles up the spread, basically assuring that all market orders fill at horrible prices.  You need leverage, micro-second trading ability, and FOK orders to accomplish this.  Bitcoin provides none, and as I said earlier, he posted that he had 5 trades.

[BitcoinPorn]

You guys really overestimate the impact of bots on bitcoin, it's minimal, and most of the impact is positive, not negative.

Adjusted title of the thread, I am still kind of confused on overall impact is all.  I keep thinking about the positives it gives by having a kind of regulation there, but how unnatural it is as well.  Still not sure how I feel.

[Tasty Champa]

dude said it wasn't him.

I'm jealous too... xD

[Tasty Champa]

You guys really overestimate the impact of bots on bitcoin, it's minimal, and most of the impact is positive, not negative.

Adjusted title of the thread, I am still kind of confused on overall impact is all.  I keep thinking about the positives it gives by having a kind of regulation there, but how unnatural it is as well.  Still not sure how I feel.

how do you have 2 different post counts?

[BitcoinPorn]

how do you have 2 different post counts?

I believe this should answer your question.

[HideousBeastManGuy]

The second bot is a joke (the link to this forum)

The pokerbotter guy nailed it.  If someone is trying to sell an algorithm it's because it isn't profitable anymore.

If you spend any real money on a system where he shows you the results from a whopping 13 trades, you deserve what's coming to you.


Don't fear the bots, I'm going to bed soon, but I'll come clean here and say that I do have a working bot that makes real money (not a ton).  I barely beat out a miner and am currently assuming nearly the same risk, although difficulty doesn't effect me.  Ask me anything you want, but I won't give any details about how my bot makes money, or how it trades... but if you want to know the logistics of making one, my opinion on how it impacts the market, who else is playing around the in the markets, I'd be happy to share.  Just not willing to give up the secret sauce.

[Tasty Champa]

how do you have 2 different post counts?

I believe this should answer your question.

http://www.bing.com/search?q=downvote+bot&go=&qs=bs&form=QBRE

[BitcoinPorn]

but if you want to know the logistics of making one, my opinion on how it impacts the market, who else is playing around the in the markets, I'd be happy to share.  Just not willing to give up the secret sauce.

I am curious.  I will gladly take this in a PM, but I prefer public as I know I would not probably be the one to ask the right questions on this subject.   Definitely interested in hearing the positives though.

http://www.bing.com/search?q=downvote+bot&go=&qs=bs&form=QBRE

So in five years the stupid phrase "there is an app for that!" will be replaced with "I made a bot for that!"

[GeniuSxBoY]

I wish I could code like that. I need someone to teach me.

[HideousBeastManGuy]

First and foremost, a bot is almost always going to perform worse than a human when it comes to discretionary Technical Analysis.  The only exception would be from HFTs (which as I explained earlier, do not exist with bitcoin due to lack of FOK).  Even then, if it were possible to slow time down to where a human could trade at the same speed as an HFT, I'd imagine the human would outperform the algorithm.  We are simply better suited to dabble in fuzzy logic.

But any strategy, be it support/resistance based, EMA based, MACD based, arbitrage based, market making based, is going to generally underperform when you automate it, compared to a human.  The most important part of using indicators in TA is to know when they are invalid and when to ignore them.  They are not perfect, they are not magic, they do not tell you anything that price isn't already saying.  To write a winning bot you have to either have a ridiculously easy opportunity, or years of trading experience FIRST.  Writing a long-run winning bot (that is 1000+ trades at a minimum) is nearly impossible with no prior knowledge of trading, with only a few exceptions.

The positives are that they increase liquidity.  If you want to sell, you can sell.  If you want to buy, you can buy.  There are also scores of losing bots, people testing bots etc.  My bot had a bug where it accidently sold for $3 less than the current Bid for a handful of coins.  I made someone's day!

The negatives only happen when they are in an environment suited for them, like we see now on most major exchanges for the "mainstream" currencies.  There have been lots and lots of power plays on real estate near very important servers for exchanges.  Basically they've created an environment where they can claim they are only using data available to the public, but there's zero chance that the public could ever compete on the same playing field.  HFTs, by definition, are finding information about supply and demand that no one else can find.

But this isn't the case with bitcoin, but I fear it might someday be.  I really don't understand why the exchanges decided to follow in the footsteps of the traditional markets and include features like dark pools.  It is unfortunate that they are actually how modern markets work, but we had/have a chance to start over again, why are we just following in the same footsteps?  You're buying access to private information.  There was a thread a few days ago where a guy was admitting that trading via the darkpool on MtGox was cheat codes, and he had manipulated the public's inability to see the correct supply and demand, to make a profit, many times.  I don't ever blame someone for taking an edge that is legally and ethically available, but I can criticize the choice for MtGox to provide such an instrument, without any counter measures that normal Dark Pool markets have like FOK orders, or some method of reporting them.  In effect, MtGox has a much much worse system than traditional markets.

But anyway, bots aren't really going to have any net impact on price that speculators don't already have.  Price goes where the market decides, and bots are controlled by market participants.  So long as you understand that just being a bot gives you no inherent advantage, other than in doesn't need to sleep, you would have to conclude that their impact on price is non-existent, the market participants want price to go there, so it does!

[BitcoinPorn]

Man, thanks for that post, read it twice.  I do see the side of bots that you are using all information that is provided and putting into a solution that is more optimized than a human can ever be (when properly set up of course, which I see is not an easy task, so many fucking variables to work with).

I am actually taking a whole new outlook regarding dark pools now.

[HideousBeastManGuy]

Re-reading my post it's a little incoherent as it was written late last night.

The main point is this:

1.  In traditional markets there are countermeasures for darkpools (FOK), we don't have that with bitcoin.  Also, most darkpool transactions are reported at some point in time in traditional markets (although, this depends on the exchange/instrument... spot Forex has major issues with this).   We don't even have this with bitcoin.

2. Bots are just an extension of regular traders, they can't do anything more than a normal trader can unless the environment is designed for them specifically.  A terrible trader will make a terrible bot.  A good trader has a chance at making a good bot (but not necessarily).  Winning trading algorithms are a subset of winning traders.  I wish I had a link, but I was reading that there's an office building in Chicago right next to the CME, that sold for north of $1B due to its close proximity to a major server in the exchange network.  HFTs require very little latency, so whoever has the lowest ping to that server, controls the HFT market.  But the moral of the story is, don't fear bitcoin bots (yet), they are never going to be as efficient as their authors, and the environment is not yet to the point where being a bot has any inherent advantage that a normal trader can't do, other than it doesn't sleep and can stay 100% focused.

3.  Fear the exchanges and darkpools.  Jered from Tradehill even chimed in on this topic and said we have every right to question what the exchange owners are doing.  DO NOT ASSUME they have your best interest in mind.  DO NOT ASSUME they know what they're doing.  To be frank, MtGox's choices with regards to how they handled the hackcrash and subsequent rewards to traders (0 commissions) demonstrates a major short-sitedness.  Zero commission for everyone creates 0 spread, but to selectively give it to some people and not others will create an unfair advantage.  Finally, DO NOT ASSUME that exchange owners don't participate in their own markets.  I believe they do, and I strongly believe they shouldn't, or at the very least need to disclose that they do.  Their participation can range from legit and honest, to downright fraud.  An example of fraud is frontrunning , which where they can place their own orders before a large order is processed and capitalize on the price movement.  When you place a trade in a market you need traders to come in AFTER you, in order to profit.  Frontrunning artificially creates that.  The owners of MtGox are the only people who see both darkpools and non darkpools, so the temptation is there.

[Synaptic]

TD;DR: Bitcoin markets and exchanges are basically shit.

OPINION: Bitcoin is basically shit.

YMMV.

[HideousBeastManGuy]

I woudn't say they're shit, it just depends on your perspective.  I think there's enough profit potential in the exchanges that I created my bot.  That would imply I don't think they are shit.  I have to hold inventory of bitcoin and USD, so I'm unfortunately forced to be a speculator when I don't want to, but I've accepted that risk for now.

The TL;DR version would be that I'm astounded at the level of trust this forum has with the major exchanges.  They just assume they'll do the right thing in the face of tremendous temptation.  I mentioned frontrunning in my last post, but MtGox could fix price at it suits them and the rest of the exchanges would follow.  That makes frontrunning look mild.  They could also come up with clever ways to have market-mulligans by rolling back trades  .

If anything is shit, it's the mentality of this forum.  That you're a hater if you question the exchanges or point out obvious flaws.  I want bitcoin to work as much as the next guy, but that doesn't mean I'm just going to ignore obvious hurdles and opportunities for unethical behavior.  As they say, absolute power corrupts absolutely, and right now MtGox has that type of power until some other exchange can draw a large share of the market.

[BitcoinPorn]

Regarding TraderTimm's comment about the reported 'net worth':
The periods in time where it appears that the reported net worth is tracking the price of BTC simply means that the bot is holding inventory, waiting for the target price to be reached. Net worth isn't used anywhere in the GA algo, it's only reported as a visual sanity check. And as i mentioned in the original thread, as the GA constantly updates to adapt to new data, the reported 'net worth' will constantly change. The system is internally weighted (within the fitness function) to favor more recent trades. Therfore, the reported net worth should not be taken as an indicator of performance of the system over time. It's was simply a snapshot in time. Also, at no time was the complete account balance in play which further diminishes it value in calculating ROI.

Regarding HideousBeastManGuy comment: "You cannot run HFTs unless you have Fill-Or-Kill (FOK) orders."

I do have that functionality! While not officially supported in the MtGox API, it took almost no effort to implement in software. I have automatic order cancellation (buy wait), stop loss, stop age, commit to sell target and sell target prices built into my trade platform.
I am limited in two ways to meet the true definition of HFT - Latency in the MtGox API & volume/volatility. Notice that neither one of them have anything to do with my system. While it can be argued that I am not currently meeting the true definition of HFT, my system IS HFT ready. Once user keys are re enabled through the websocket, that only leaves volume/volatility as the limiting factor.
That being said, the bot currently averages 100 trades a day in a throttled state (used for debugging) with about 600 total trades.

He also said: "Also, I'm not sure I subscribe to the whole genetic algorithm being something that can be applied to markets, as it seems to me you'd have a massive hurdle in that the algorithm would want to choose the best winners on a curve-fitting basis. Perhaps he's taken measures to avoid this"

Simpleton fitness functions will always curve fit. It's funny how computers do exactly what you tell them. This is why, in my original post, I talked about the difficulty in coming up with a suitable fitness function.
This is probably why 90% of people who will / or ever have tried to tackle GAs for market trading have or will give up. If you ever hear of anyone trash GAs as being limited or worthless, I'd bet money it's because they couldn't develop a suitable fitness function.

At this point I could probably write a book called 'Fitness Functions: I bet your doing it wrong' with everything I've learned.

Anyway, yeah I have taken multiple measures to avoid curve fitting. I have a power function for non linear age weighting, an exponential function for non linear trade period weighting and the latest hack is market variance detection (True Range) run through a histogram to split the GA system into variance quartiles. Simulated buy orders are now gated to the quartile linked gene being tested. Now I can optimize for multiple market conditions in parallel and switch active strategies on the fly dependent on current market conditions. There are a couple of other key optimizations I made to avoid curve fitting, but this post is getting too long.

I do however agree with this: "Writing a long-run winning bot (that is 1000+ trades at a minimum) is nearly impossible with no prior knowledge of trading, with only a few exceptions."

I'm not the exception, I did have 'prior knowledge'.


http://www.reddit.com/r/Bitcoin/comments/ipr35/bots_keeping_the_price_still/c25v66k

[Leon]

Man, thanks for that post, read it twice.  I do see the side of bots that you are using all information that is provided and putting into a solution that is more optimized than a human can ever be (when properly set up of course, which I see is not an easy task, so many fucking variables to work with).

I am actually taking a whole new outlook regarding dark pools now.

+1

[Synaptic]

I woudn't say they're shit, it just depends on your perspective.  I think there's enough profit potential in the exchanges that I created my bot.  That would imply I don't think they are shit.  I have to hold inventory of bitcoin and USD, so I'm unfortunately forced to be a speculator when I don't want to, but I've accepted that risk for now.

The TL;DR version would be that I'm astounded at the level of trust this forum has with the major exchanges.  They just assume they'll do the right thing in the face of tremendous temptation.  I mentioned frontrunning in my last post, but MtGox could fix price at it suits them and the rest of the exchanges would follow.  That makes frontrunning look mild.  They could also come up with clever ways to have market-mulligans by rolling back trades  .

If anything is shit, it's the mentality of this forum.  That you're a hater if you question the exchanges or point out obvious flaws.  I want bitcoin to work as much as the next guy, but that doesn't mean I'm just going to ignore obvious hurdles and opportunities for unethical behavior.  As they say, absolute power corrupts absolutely, and right now MtGox has that type of power until some other exchange can draw a large share of the market.

Son, you have to realize that most of the people here are mouth-breathers who found a vehicle for their impotent societal frustrations. They basically lack the critical thinking ability to understand why the bitcoin exchanges are HILARIOUSLY corrupt ON PRINCIPAL and certainly are not in their best interests to use. Especially Gox.

The majority of the market is miners selling to speculators.  The miners don't care about the trustworthiness of the exchanges because they're just convenient BTC to USD machines.  The speculators are generally just amateur day-traders who aren't intelligent or affluent enough to play in a real market like FOREX, and will use whatever is available to then, trustworthy or not.

And the zealots that foresee a glorious future for bitcoin?  Well, generally cult members lack the cognitive capacity to understand their plight, despite the protestations of those around them.