Trend following system only profits in 2008???

[Adamus]

I'm looking for a trend following system and started off looking at a 20 day MA with entries given by RSI.

It was nice'n'sleazy, if the MA was trending up, it buys whenever the RSI says it's oversold, and vice-versa, and exits with a trailing stop.

I had to play around a little with the overbought / oversold levels, in fact I set them both to 50! But the results across 10 years of daily data in 25 markets showed that it was profitable in 18 markets.

I then looked more closely - it was only 2008 that was profitable! Seems ridiculous! Crazy! 13 out of the 25 markets made massive profits in 2008 and otherwise were pretty much loss-making.

Is this the effect of the credit crunch?

More to the point, is this something that affects all trend following? Should I leave out 2008 when experimenting and searching for other set-ups?

I already knew my short-term break-out strategy was very profitable in 2008/9 but I wasn't expecting it to be even more pronounced like this for longer term trend following.

 

[meanreversion]

Without knowing all your trading rules and money management, it's hard to comment - it's really for you to dig a little deeper. What I would say is that this isn't a pure breakout strategy... you're waiting for dips in an uptrend to buy... what if you never get the dip and miss the move?

The last few years have seen some monumental trends.. JW Henry returned 80 pct or so in 2008 for example.. I think the average return in 2009 was more like 5-10 percent.

Read 'The Way of the Turtle' by Curtis Faith for a detailed explanation of a complete trend following trading system.. code it up if you're able and play around with it. You'll find that some markets (FX) are more receptive to trends, whilst some not so much (S&P 500).

 

[Adamus]

I read the 'Turtle Rules' PDF by Curtis Faith, which was pretty clear. I may just code that up, although I would guess their 'always-in' methodology leaves them open to bigger drawdowns that my little $50K account size can handle.

Reading between your lines, it looks like I should leave out 2008 entirely. And I'll leave out 2009 and keep it for walk-forwards. Got some great advice to use random windows for walk-forwards, but that would require a whole new configuration for my software.

My money management is not applied to these results - they are all the result of one lot in each futures market. For money management, I'll allocate a % of my account to each system I trade and take positions based on optimal f. (Not pure optimal f as Ralph Vince described it, but a safer variation of it).

I see what you mean about the disadvantage of missing out on a trend completely by missing the entry, but off the top of my head, I figure I wouldn't be getting much slippage if I entered on a limit order. With the previous break-out strategy I played with, I held back from trading it mainly because I thought I'd get killed on slippage. It's quite high frequency and I was just factoring in a tick for slippage and a tick for the spread and then adding the commission. So for the German Bund I'd use transaction costs of €31.

Entering on limit orders would mean less I figure, and then also the trend following system's average profit per trade should be far far more than the transaction cost - the compromise being a poor win:loss frequency ratio.

Am I making sense? Probably only to myself :rofl:

 

[meanreversion]

Backtesting over the last 10 years, even 5 years, includes every type of market you'll ever encounter (extreme low vol, extreme high vol, stuff in between) so do NOT exclude 2008. In fact, I can't think why you would ever selectively dis-include years from your testing.... test as much as you can.

Position size should relate to market volatility, as outline in 'Turtle' and also by Van Tharp, i.e. you should risk a fixed amount per ATR move.

You miss my point about limit orders. A breakout system buys new highs or sells new lows. If you're using limits waiting for some kind of pullback, you might never enter the trade and completely miss the move. Why not combine a pure breakout system with one that also buys dips/sells rallies in the direction of the overall trend.

There are no shortcuts here. If your plan is to trade mechanically (and that can work, many CTAs and hedge funds do) then you need 110 pct belief in your system. And you will only get that from coding it yourself and testing it to oblivion. If you believe in your system and that it has positive expectancy, it will enable you to trade through the inevitable periods of drawdown.

You can make money off $50k but the faster you try to do it, the greater risk of ruin you run. Typically, a trend system which makes (e.g.) 30 pct a year will suffer 30 pct drawdowns. So if you're looking to make 50 pct a year, can you really stomach the potential 50 pct drawdown?

Don't underestimate the power of compounding.. if you make 20 pct a year then after 10 years you have 6 times what you started with. Trading is a medium to long term business, don't try to double your money overnight.

 

[Adamus]

do NOT exclude 2008

OK good point. But just excluding 2008 makes it easier to discover a certain type of system when I'm reasonably sure that this type of system is always going to be profitable in 2008. I mean, if I leave it in, the 2008 profits will swamp and hide the PnL from the other years. Or at least that what just happened.

Regarding your point about annual return, drawdown and risk of ruin, that's certainly something that I've really got to get a grip on. I have a vague idea about it but I'm trying to find a system that looks reasonable first. Cart before horse maybe.

So far, found nothing. Still learning though.

 

[Rossini]

''I then looked more closely - it was only 2008 that was profitable! Seems ridiculous! Crazy! 13 out of the 25 markets made massive profits in 2008 and otherwise were pretty much loss-making. ''


Seems to me that you should reverse your system. Build a structure around it that would save you when a year like 2008 happens, if you did this then would the profits for all the other years be acceptable?

 

[Adamus]

Hi Rossini,

unfortunately it is transaction costs that make it unprofitable. Generally speaking I aim for large numbers of trades when given the option and I simulate with commission + 1 tick for spread + 1 tick for slippage.

I found that my system just loves 1999 / 2000 as well. I've no idea what was going on at that point in time, I think I was backpacking around S.E.Asia. Perhaps I should go and do that again :jester:

 

[Rossini]

Lol, 99/00 was the bursting of the dot com bubble. Can you find a way to trade less?

 

[meanreversion]

Adamus, any more luck in your research?

 

[Adamus]

I abandoned end-of-day data to focus on hourly bars.

Getting a lorry-load of tick data and trying to import it into my software caused havoc with my software setup. I ditched TradeStation for NinjaTrader and I'm not quite back to where I was - i.e. I'm still importing data (5 to 10 gigs per symbol) and learning NinjaScript - their order management algorithms are slightly more convoluted than TradeStation's.

I am on the verge of the really important bit - running the optimizations and walk-forwards. Then I'll know (a) whether this strategy makes money on intra-day bars too (b) how much capital I'll need (c) how much rent money I'll need before I make a profit.

 

[meanreversion]

How's that going, have you found something useful.. I would think trend following is more of an end of day type pursuit, but technically one needs trends at whatever timeframe one trades...

 

[Adamus]

Funnily enough the 60-min bar system I'm looking at right now has an equity curve that is exactly the same as the best equity curve I could find with EOD data.

[FONT="Courier New"]
Total Net Profit			$47715.00
Gross Profit			$404825.00
Gross Loss	        		$-357110.00
Commission			$6275.00
Profit Factor			1.13
Cumulative Profit		$47715.00
Max. Drawdown			$-26645.00
Sharpe Ratio			0.14
	
Start Date				01/07/1998
End Date				30/06/2009
	
Total # of Trades		1255
Percent Profitable		37.13%
# of Winning Trades		466
# of Losing Trades		789
	
Average Trade			$38.02
Average Winning Trade	$868.72
Average Losing Trade	$-452.61
Ratio avg. Win / avg. Loss	1.92
	
Max. conseq. Winners	7
Max. conseq. Losers		17
Largest Winning Trade	$14750.00
Largest Losing Trade		$-2680.00
	
# of Trades per Day		0.31
Avg. Time in Market		1.99 days
Avg. Bars in Trade		34.2
Profit per Month		$362.56
Max. Time to Recover	1667.04 days
[/FONT]

As a start, I guess I'm happy with it. Hopefully later today or tomorrow I'll get something better. However my cynical self says this depth and length of drawdown is unavoidable.

On EOD data, if I added a market, because the theory said my drawdowns should decrease. I always found my drawdowns stayed the same. The profit increased, but profit is secondary compared to the size of these drawdowns.

Actually I know why that is the case - the markets I add together into a portfolio are based on the same system, just different optimisation so I guess they're heavily correlated.

 

[forexkunta]

Funnily enough the 60-min bar system I'm looking at right now has an equity curve that is exactly the same as the best equity curve I could find with EOD data.

[FONT="Courier New"]
Total Net Profit			$47715.00
Gross Profit			$404825.00
Gross Loss	        		$-357110.00
Commission			$6275.00
Profit Factor			1.13
Cumulative Profit		$47715.00
Max. Drawdown			$-26645.00
Sharpe Ratio			0.14
	
Start Date				01/07/1998
End Date				30/06/2009
	
Total # of Trades		1255
Percent Profitable		37.13%
# of Winning Trades		466
# of Losing Trades		789
	
Average Trade			$38.02
Average Winning Trade	$868.72
Average Losing Trade	$-452.61
Ratio avg. Win / avg. Loss	1.92
	
Max. conseq. Winners	7
Max. conseq. Losers		17
Largest Winning Trade	$14750.00
Largest Losing Trade		$-2680.00
	
# of Trades per Day		0.31
Avg. Time in Market		1.99 days
Avg. Bars in Trade		34.2
Profit per Month		$362.56
Max. Time to Recover	1667.04 days
[/FONT]

As a start, I guess I'm happy with it. Hopefully later today or tomorrow I'll get something better. However my cynical self says this depth and length of drawdown is unavoidable.

On EOD data, if I added a market, because the theory said my drawdowns should decrease. I always found my drawdowns stayed the same. The profit increased, but profit is secondary compared to the size of these drawdowns.

Actually I know why that is the case - the markets I add together into a portfolio are based on the same system, just different optimisation so I guess they're heavily correlated.

Try adding a cross of the MACD 0 line as an additional confirmation.
Kent