Topic: Goomboo's Journal - page 15. (Read 281462 times)
Interesting. How much of your capital do you usually have in play? The rule I generally hear bandied about is that you shouldn't have more than 2% of your capital invested in any single trade, and you shouldn't have more than 20% of your capital at risk.
How do you avoid whipsaws with your indicators?
You can mitigate whipsaw conditions by waiting for a new high or low to be established prior to entering a new trade. But I've found that I've been better off simply toughing out the whipsaw periods rather than adding a "filter" to the system. Also, I've found that on the higher timeframes I experience less whipsaw and more trend.
- bad time frames :the time frames < hourly give bad results, so M5 isn't the best choice... and 100 days is a bit short for daily time frame
- the tested period is difficult : price goes from 140 to 105 without long rising trend...
-> Earning some profit in these conditions is good enough
- the tested period is difficult : price goes from 140 to 105 without long rising trend...
-> Earning some profit in these conditions is good enough
Thank you very much for this deep insight
(2) why the extreme gap in time intervals? 1d is pretty slow moving, probably *too* slow moving for this market. 5min is almost certainly way too fast. Conventional wisdom would dictate to try 1h, maybe 2h time intervals.
I actually test on 7 different candles, these were just to two I posted. Given what Marcus posted:
I am not sure my charts are ready for prime time, but that is why I posted to see what I was missing.
The problem is that anything before ~Apr-18 and your results are skewed by the 'bubble'. I am looking for more 'real world' results given that any of these signals would trip during a bubble like that anyway.
In my backtesting the bubble seems to make surprisingly little difference I personally don't like the indicator that much. It's too much on one screen :p I prefer to only have a few simple indicators on the screen at a time.
What I like most about Ichimoku is it identifies when you're entering and exiting a consolidation period, so you can avoid whipsaws. How do you avoid whipsaws with your indicators?
So I wrote some script to use the Gekko Bot to do batch-range backtesting to generate charts similar to what Goomboo posted back in Feb 2012.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
Nobody reacts to such scaring low profits ?
- bad time frames :the time frames < hourly give bad results, so M5 isn't the best choice... and 100 days is a bit short for daily time frame
- the tested period is difficult : price goes from 140 to 105 without long rising trend...
-> Earning some profit in these conditions is good enough
So I wrote some script to use the Gekko Bot to do batch-range backtesting to generate charts similar to what Goomboo posted back in Feb 2012.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
mtgoxUSD-D1
mtgoxUSD-M5
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
mtgoxUSD-D1
Code:
start time: 2013-04-27 18:00:00
end time: 2013-08-05 18:00:00
timespan: 100 days
end time: 2013-08-05 18:00:00
timespan: 100 days
mtgoxUSD-M5
Code:
start time: 2013-04-17 18:50:00
end time: 2013-08-06 08:30:00
timespan: 110 days
end time: 2013-08-06 08:30:00
timespan: 110 days
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
Very nice. Thanks for sharing. It's pretty high on my own to-do list, write a similar program. Problem is, I've never worked with the mtgox? bitcoincharts? APIs before, so that'll need some time to get used to... funny enough, I'm trying to convince my gf who has a programming project due in a month to do this for me, for her to get credit, for me to earn sweet sweet bitcoin
Anyway, looks good to me, but two questions:
(1) why don't you backtest longer (say, 1 year back) but *exclude* the bubble data, that one could consider an outlier. Say, you perform the test on [August 1st 2012 to February 1st 2013] and [May 1st 2013 to now].
Not the perfect solution, but it's not entirely unheard of to declare some data points outliers and ignore them
(2) why the extreme gap in time intervals? 1d is pretty slow moving, probably *too* slow moving for this market. 5min is almost certainly way too fast. Conventional wisdom would dictate to try 1h, maybe 2h time intervals.
So I wrote some script to use the Gekko Bot to do batch-range backtesting to generate charts similar to what Goomboo posted back in Feb 2012.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
Nobody reacts to such scaring low profits ?
A question for you: what do you think of the Ichimoku indicator?
I personally don't like the indicator that much. It's too much on one screen :p I prefer to only have a few simple indicators on the screen at a time.
I would normally agree with that (I hate a busy chart), but when color coded, it looks pretty clear.
It's meant to be "At a glance" and for that you need a few lines that indicate the trend and the state of that trend. Then it's only the relation between the green/red lines, the price to the green (conversion) line, and the price to the cloud.
That is interesting. I wonder if there is a bug in Gekko.
I will look into this tomorrow.
I will look into this tomorrow.
A question for you: what do you think of the Ichimoku indicator?
I personally don't like the indicator that much. It's too much on one screen :p I prefer to only have a few simple indicators on the screen at a time.
Looking at one of the most profitable combos, the 6/50, you only get one sell and one buy signal during that time frame. I think you need a longer test period to really validate.
I actually got 5:
Code:
2013-08-07 09:02:24 (INFO): I'm gonna make you rich, Bud Fox.
2013-08-07 09:02:24 (INFO): Let me show you some Exponential Moving Averages.
2013-08-07 09:02:24 (INFO): Preparing backtester to test strategy against historical data.
2013-08-07 09:02:24 (INFO): Profit reporter active on simulated balance
2013-08-07 09:02:24 (INFO): Calculating EMA on historical data...
2013-08-07 09:02:24 (INFO): calculated initial EMA, simulating remaining candles
2013-08-07 09:02:24 (INFO): ADVICE is to SELL @ 144.000 (-14.790) at 2013-04-28 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to BUY @ 98.100 (0.643) at 2013-05-02 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to SELL @ 123.500 (-0.978) at 2013-05-16 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to BUY @ 111.000 (1.849) at 2013-06-06 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to SELL @ 107.960 (-2.621) at 2013-07-29 18:00:00
2013-08-07 09:02:24 (INFO): WARNING: BACKTESTING FEATURE NEEDS PROPER TESTING
2013-08-07 09:02:24 (INFO): WARNING: ACT ON THESE NUMBERS AT YOUR OWN RISK!
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) start time: 2013-04-27 18:00:00
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) end time: 2013-08-05 18:00:00
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) timespan: 100 days
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) start price: 93.07
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) end price: 106.76002
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) Buy and Hold profit: 14.709%
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) amount of trades: 5
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) original simulated balance: 2.500 undefined
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) current simulated balance: 4.006 undefined
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) simulated profit: 1.506 undefined (60.259%)
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) simulated yearly profit: 5.363 undefined (214.524%)
2013-08-07 09:02:24 (INFO): Let me show you some Exponential Moving Averages.
2013-08-07 09:02:24 (INFO): Preparing backtester to test strategy against historical data.
2013-08-07 09:02:24 (INFO): Profit reporter active on simulated balance
2013-08-07 09:02:24 (INFO): Calculating EMA on historical data...
2013-08-07 09:02:24 (INFO): calculated initial EMA, simulating remaining candles
2013-08-07 09:02:24 (INFO): ADVICE is to SELL @ 144.000 (-14.790) at 2013-04-28 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to BUY @ 98.100 (0.643) at 2013-05-02 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to SELL @ 123.500 (-0.978) at 2013-05-16 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to BUY @ 111.000 (1.849) at 2013-06-06 18:00:00
2013-08-07 09:02:24 (INFO): ADVICE is to SELL @ 107.960 (-2.621) at 2013-07-29 18:00:00
2013-08-07 09:02:24 (INFO): WARNING: BACKTESTING FEATURE NEEDS PROPER TESTING
2013-08-07 09:02:24 (INFO): WARNING: ACT ON THESE NUMBERS AT YOUR OWN RISK!
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) start time: 2013-04-27 18:00:00
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) end time: 2013-08-05 18:00:00
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) timespan: 100 days
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) start price: 93.07
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) end price: 106.76002
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) Buy and Hold profit: 14.709%
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) amount of trades: 5
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) original simulated balance: 2.500 undefined
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) current simulated balance: 4.006 undefined
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) simulated profit: 1.506 undefined (60.259%)
2013-08-07 09:02:24 (INFO): (PROFIT REPORT) simulated yearly profit: 5.363 undefined (214.524%)
But I agree about the time period. The problem is that anything before ~Apr-18 and your results are skewed by the 'bubble'. I am looking for more 'real world' results given that any of these signals would trip during a bubble like that anyway.
Looking at one of the most profitable combos, the 6/50, you only get one sell and one buy signal during that time frame. I think you need a longer test period to really validate.
So I wrote some script to use the Gekko Bot to do batch-range backtesting to generate charts similar to what Goomboo posted back in Feb 2012.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
mtgoxUSD-D1
mtgoxUSD-M5
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
I ran multiple candles / timeframes / values. Here is some candy (timeframe picked to be after the Apr crash):
mtgoxUSD-D1
Code:
start time: 2013-04-27 18:00:00
end time: 2013-08-05 18:00:00
timespan: 100 days
end time: 2013-08-05 18:00:00
timespan: 100 days
mtgoxUSD-M5
Code:
start time: 2013-04-17 18:50:00
end time: 2013-08-06 08:30:00
timespan: 110 days
end time: 2013-08-06 08:30:00
timespan: 110 days
The numbers you see is the expected PROFIT. The simulation started with a modest 2.5 BTC. Obviously <0 is not good (a loss).
Interestingly enough it looks like M5 candles are NOT a good choice right now.
Spoke too soon, I've got it now. Thanks again. Very useful.
Beautiful charting site, thanks. But I can't figure out how to set the time frame, only the length of time showing on the screen.
Unfortunately I know of no charts for Bitcoin which show Ichimoku.
https://www.tradingview.com/e/fWyEQutH/
(4h Ichimoku cloud in blue, 1d red and 2d orange)
Goomboo, thank you for taking the time to give us an education with this journal of yours. Thank you especially for insisting on the importance of backtesting. I now feel as if I have something solid on which I can base my trade decisions. Backtesting has become my new hobby!
A question for you: what do you think of the Ichimoku indicator? I've done lots of backtesting with various currency pairs and have gotten good results. Each pair has a different personality, of course, so you have to use Ichimoku in a different way with each pair. Unfortunately I know of no charts for Bitcoin which show Ichimoku.
Respectfully,
Marcus
A question for you: what do you think of the Ichimoku indicator? I've done lots of backtesting with various currency pairs and have gotten good results. Each pair has a different personality, of course, so you have to use Ichimoku in a different way with each pair. Unfortunately I know of no charts for Bitcoin which show Ichimoku.
Respectfully,
Marcus
Jump to: