Topic: Chrome Browser extension: MtGox trading bot - page 3. (Read 72509 times)

I've simulated the bot's behaviour, backtesting MACD strategy, in order to find the best threshold values for maximizing the profit.
Test goes from May 3rd to June 6th.
Seems like positive results are impossible with sampling times inferior to 1 hour. Additionally, best threshold values in this case are threshold-buy=0.6%, threshold-sell=-0.4% so ca. 10 times the recommended value of 0.05 .
This bot is not suitable for fast trading!

So ive been using the default settings, but for what ever reason it has been selling low and buying high  what could the problem be? It's not like crazy ivan is on or anything

Threshold have absolutely nothing with this. Threshold defines only relative power of trend.  Weak trend make profit at the end of long period. If you have high threshold, you loose it.
On other side - powerful trend on short interval easy force you to buy, but in the moment you really buy, price is changed because you make decision on basis of average price value .
Imagine letter V on price curve. Average value will be -0.33 , it's value of decision. But value of trade will be at the end of interval / of V - i.e 0. This error is a price of using average values.
You can get big price difference on low threshold, and you can get 0 on high.   

It's a simple math...

Trend threshold say: "there is good conditions for trade, as good as your settings are". It newer say "you will get 1%(2..3..100%) of profit". You easily lost all your money with any bot on any strategy, if you don't use simple loss protection.  

strange: why did not buy?



edit: never mind. It was a Gox glitch: for a while the bot reported a "parsement problem", then it bought.  

That is where the "Treshold" parameter was supposed to come handy.

Unless I misunderstood your problem?

I like this bot and I like style it made. Thank you.  

But IMHO bot have a fundamental problem.

As soon as Gox eats 0.6% of every trade, bot ( with any settings ) should have trade protector with minimum at 1.2 % of trade price + minimal profit.
Values and Settings of EMA/MACD have nothing with this. Them concerning, say, "integrated fraction" of price curve, and shows mostly a trend strength.

Bot should sell/buy only when the price difference within the last trade cross protection limit. Gox is not a forex, where you can trade on .000.... % and still have a profit.

But bot trade on on basis of current average price, and we newer know what is the real price at the moment we make order or what is the real current price. So, something should be done to know real trading price and to protect next trade on basis of previous. Very first trade, to be done , additionally needs manual trade button or some other solution.  

Resolving this problem is a condition of using any bot with any algorithm. Like a powered computer.


edited::

Awesome stuff man, donation sent

I must say I'm impressed... Will take me some time to figure out good settings but it looks good. I'm making second account on my girlfriends name for MACD strategy bot... Thanks!

And if you have time for a baby number 3. This looks interesting...
https://bitcointalksearch.org/topic/m.2330792
https://bitcointalksearch.org/topic/m.2340317
https://bitcointalksearch.org/topic/m.2341374

Good idea rikigst...

EDIT: Would it be possible to run multiple strategy from a single bot to multiple accounts?

Yes, "Crazy Ivan" just calls the sell-function in MtGox instead of the buy-function and vice versa. So if you tweak the "buy-settings", they will be used to trigger a sale and vice versa. And it works for MACD as well...

thanks for your work on this. I just had an idea about how The Crazy Ivan could be used. does it use whichever settings you have set in the trade settings but just sells on the buy and vice versa?

Bravo! I just installed it, seems working. Maybe a 0.1% buy threshold is safer for 1 hour intervals.. Keep up the good work Tobli!

Awesome work! Will look at your MACD implementation when I have some time.

Hi all!
I've been working on a new version of the bot for a while, now including MACD indicator/strategy!!
It's not totally finished (mostly help/info-functions that needs to be fixed), but I have been running it in MACD mode for about a week, and I think the logic, the charts and the trading all work as it should!

As I think we will get our second baby today(!), I'm publishing it for you to play with, even if it's a bit rough.
I have changed it's name and put it in a new repository:
https://github.com/TobbeLino/TobliBot
(when it's finished I'll publish it in the Chrome Store)

A few notes regarding the MACD-strategy:
- The standard EMA-values for the MACD-calculation is a bit different from Goomboo's simple EMA-strategy.
It's normally 12/26 (instead of 10/21 for Goomboo's). And the MACD-indicator also has a signal line that is per default a EMA(9) of the MACD-line. Clicking the link "strategy defaults" after changing bot-strategy loads the default parameters for each strategy.

- The EMA-thresholds should be a lot lower than for Goomboo's simple EMA-strategy.
I have made the MACD-diff values (that are used as the indicator) relative rather than absolute by dividing with the long EMA (according to how PPO is calculated: http://stockcharts.com/school/doku.php?id=chart_school:technical_indicators:price_oscillators_pp but this bot does not multiply it by 100 as the PPO) so the thresholds should be constant regardless of the current price range.
I have empirically chosen 0.05% as the default values for buy/sell thresholds in MACD-mode, but feel free to experiment!

And (as also written on the new options page):
If my bot makes you rich, please consider donating a fraction of your profit!
...or, of course, also if you appreciate my work for other reasons...
Bitcoin: 1LUqdAXvH9gbYemZKeiMrVJ5njhm6ZvKmF



Cheers!
/Tobbe

Pot meet kettle. Didn't assume you endorsed it. Just made a light hearted jest, at a rather out of place example being used. Sensitive much?

It appears, you have mistaken what I wrote with me giving some sort celebrity endorsement to fapturbo.  I implore you to reread what I previously wrote or quit reading into what is written into a post, since clearly I did not endorse fapturbo nor did i use any of my celebrity to imply anything about fapturbo. 

I do agree that you shouldn't always trust everything, especially things with a download button. However:

If you are a programmer why you don't you just read the code yourself instead of not trusting the maintainer? As a programmer you are always building on top of / using stuff others created, it is just not feasible to reinvent the wheel every time imo. I think open source software like this needs more verification from people able to read the code.

I don't quite understand your position: have you read the source of bitcoin-qt? Do you even use proprietary software? I know there is a difference between software of an established company and a github repo. The bot is up for over a year, the code has beed tested and read.

I don't see any real world way for the maintainers to exploit an unoptimized version of the bot: If a 1000 people were using this bot with all 1BTC on their account this would in no way be able to move a market with a size of Mt. Gox to anything which can be easily exploited. Even if there was something like a hidden button the maintainer could press to make all the bots sell or buy (which there wasn't last time I checked).

It is always important to check free tools like these that are being offered but writing things from scratch is no option except when you are a programmer yourself AND have the time to take on a new project.

I know. I'm not saying that it's sabotaged, I'm saying that it's probably an inferior version of what the creator privately uses for himself. The trading deciding algorithm can be unoptimized, or optimized. Both being legitimate trading algorithms, but one resulting in a less favorable result than the other. Resulting in scenarios that are easily exploitable by the optimized trader. I'm a programmer myself, so I know this to be a possibility, which is why I don't trust it.

If you can understand the code it's best to go out and create your own private trader.

This has nothing to do with your bot but I was trying to download your new version, 0.2.2.3, and it kept returning a 502: failure. Any other way to download your bot without having to click "zip"?

If I get the time I'll try to implement some experimental functions that takes the price you paid into account, e.g. "don't sell with loss".  And maybe some kind of stop loss (need to figure out how it's supposed to work together with the other bot-strategy though, as the stop loss logic and the bot-strategy will sometimes conflict)

But right now I'm working on a different feature in the bot...

That's the reason why I build Gekko. I personally use it on my VPS which is on 24/7 anyway (for a couple of bucks a month). But you can also run it on a raspberry or similar. Gekko doen't have the overhead of the whole browser Chrome (doesn't even need gui, runs fine on my ubuntu-server), etc. Also it's extremely easy to daemonize / keep running for days.