Equity curves - how smooth do you like yours to be B4 you consider trading it?

[mr_cassandra]

Thanks for your input

Granted a backtest guarantees nothing, how much trading in real-time with the same algorythm before you would feel confident ?

How about max draw during an uncompleted trade?

Thats a very good question. If you include max drawdown during the trade, this will be to make sure you wont get below the margin required by your broker. If not included, it will indicate the risk profile of the algorithm it self.

IE if the system was generating X% annually I could tolerate an intratrade draw of y%

You are talking about anually - please note that you can not base your automated trading of algorithms trading a few times a week:

1) This timeframe is too short. The way prices are moving on the stock exchange are changing, an algorithm which worked last year may not work this year.

2) Its close to impossible to backtest such algorithm, since you need to backtest on 3-8 years of historical data, thus on historical data which may not work any more.

3) Before you trade for real, you have to trade simulated using live data feed, but on a paperaccount. Thus you have to test simulated during a year - which is not operational.

4) Take care of public algorithms, they are posted since they didnt work for the developer. Especially be alerted on bakctests using daily bars, trading a few times a week.

 

[JTrader]

The smoothest & most diagonal of equity curves are likely to flounder at some stage. Its good to spend some time thinking through various what if scenarios within your own trading framework - planning what you will do & at what stage etc.

 

[TickCOM]

Granted a backtest guarantees nothing, how much trading in real-time with the same algorythm before you would feel confident ?

Dont think there are any rules for this, I suppose this depends on the trader, the algorithm and the timeframe traded.

Personally, I want a minimum of 100 roundtrips preferable 200 roundtrips. My own personal algorithms all trades 4-8 roundtrips pr day - thus papertrading 1-2 months. During the papertrading, I compare the behavioure of the papertraded algorithm with the backtecting. How the trades are divided over the day and the week, gain/loss profile etc.

I trade mainly the GLOBEX currency futures, since they can be traded 23h a day, have a 1 tick spread, are ultra liquid.

Have 3-4 potential algorithms running for papertrade during the week. When on holiday, I always make sure there is a LAN connection in the room, and bring the portable with me :cheesy:

 

[Crap Buddist]

Happy xmas everbody......

I found this tip on back of fag packet. I share.


Whop Moving average on your current equity graph until desired visual smoothness looks good. Make note of the average number multiplier thats gives good visual smoothies.

Divide your current bet size by this number n chill.

eh voila.

Thats the way the cookie crumbles.


And your equity be smoother than The Fonz...... Heyyyyyy!!!

 

[the blades]

eh voila. And your equity be smoother than The Fonz...... Heyyyyyy!!!

The Fonz didn't play any musical instruments, did he?

seriously, though it would make most sleep at night, wouldn't it render the trading process pointless - ie your bets would be that small you'd make no money?

I haven't tried it mind.

Personaly I find trading numerous systems gives an overall equity curve that's tradeable, even though some of the components wouldn't be by themselves.

UTB

 

[Crap Buddist]

The Fonz didn't play any musical instruments, did he?

seriously, though it would make most sleep at night, wouldn't it render the trading process pointless - ie your bets would be that small you'd make no money?

I haven't tried it mind.

Personaly I find trading numerous systems gives an overall equity curve that's tradeable, even though some of the components wouldn't be by themselves.

UTB

Maybe peeps can use it as a test/measure of method used, whats the multiplier needed to make my equity smooth ? <10 or >10 ?

 

[Mathemagician]

I have found that maximum drawdown carries little information about the viability of a trading model. Essentially, sampling variation renders the maximum drawdown seen in any series of trades effectively meaningless. For a discussion of the basis for this conclusion and an illustration of the surprising variability in outcomes that are be observed even under the most ideal conditions, take a look at the article and spreadsheet here. When I performed this analysis initially, I was quite shocked at the results. It's kind of discouraging, actually, until you realize that there are statistical techniques for quantifying this variability and these translate into valuable decision tools for trading and setting expectations.

jj

 

[zentrader22]

@Mathemagician,

quite shocked, really?

But you have only resampled the original backtest ratios and it's normal that you get some higher DDs...but that's only Step 1.

If you process also Step 2 (data scrambling or data simulation, new back tests with simulated synthetic data, new resampling or monte carlo based system simulation) you get even more alternative future scenarios and possible far higher DDs...

I've described this method in Traders' some months ago:
http://www.zentrader.de/mcs_article_traders2007.pdf

bye,
zentrader

 

[Mathemagician]

Excellent article!

No, I do not resample in-sample data under any circumstances (not even as step 1, since it is really a pointless exercise). For models with a limited trading history I resample only Walk-Forward Optimized data. For established models I resample live results. FWIW, I prefer bootstrapping with a modification that preserves serial dependency to Monte Carlo.

jj

 

[zentrader22]

...yes, this "Step 2" area is really painful, but only there you can see some of the "black swans" right in time...

Amazon.com: The Black Swan: The Impact of the Highly Improbable: Books: Nassim Nicholas Taleb

bye,
zentrader

 

[Samuel_J]

The smoother the curve, the more delayed the trading signals will arrive. There are no really good ways around this, I'm afraid.

But, you can improve the smoothing a bit by making the smoothing more intelligent than simply calculating an average of the last x days...there are several adaptive smoothing filters out there:

Two of the most advanced are:
Moving Average Plugin:
Jurik's JMA

Trading Software with improved moving averages:
Optimal Trader's OptAMA

Some trading stations have simpler adaptive moving averages like the Kaufman's Adaptive Moving average included....

Expect no miracles from adaptive moving averages, but they do improve indicators a bit.

With kind regards,

Samual

 

[nobel_1101]

I look at drawdown in three ways.

1. drawdown 'during' a trade
2. drawdown from a completed trade
3. drawdown from worst string of losing trades

Which would you say is most meaningful ?

Hi,

This is a very interesting question. I've asked myself a similar question. Could you not consider 1 and (2 & 3) as separate and meaningful in the their own right?
(1) - Is just a measure of how good your system is at capturing profit during a trade. As long as your initial risk is acceptable this might not be too important (unless it's giving away shed loads!)
(2) & (3) - I think both of these could be considered position sizing issues. Although I think (3) is more informative as it also includes drawdown from multiple trades.