[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

More on perpetual data



PureBytes Links

Trading Reference Links

I agree with Mark (big shock :-) and Tim that perpetual data may give
better signals and cleaner chart patterns. What follows isn't for them.
I know those guys understand this stuff inside and out. 

The problem is with using perpetual data as data 1 in TS and backtesting
it in a system. I submit that, if you want to get your signals from
perpetual data, it should be data 2. For accurate backtesting stats, you
should enter your trades (data 1) either on real data or, for a long
backtest, on a spread adjusted continuous contract.

Here's the proof. The perp closing prices are from the file on Mark's
web page. The spread adjusted closes are from one of my homemade files.
The DSP numbers are BMI settlement prices. Mark's data only went through
9/98 so I did a simple system that bought the S&P futures on 1/7/98 and
held for 3 months with one rollover. 

Date	DSP9H	DSP9M	Perp	Sprd Adj
1/7/98	974.00	984.00	975.85	1032.30
2/11/98	1023.70	1034.40	1029.65	1082.00
3/11/98	1068.60	1080.10	1078.85	1126.90
4/8/98		1111.80	1114.00	1158.60

As you can see from the results below, the perpetual data overstates
profits by about 12 handles. Depending on the system, it might overstate
or understate profits but you can be sure they will be wrong. The spread
adjusted data, while perhaps no good for generating signals, gave an
accurate profit number.

Saying it again, perpetual data is fine for generating signals and
plotting on charts. But it should not IMHO be used in TS as data 1 when
backtesting a system. Make it data 2 instead.

Test 1: perpetual data

	1/7/98 buy	 975.85
	4/8/98 exitlong	1114.00
	profit		 138.15

Test 2: spread adjusted data

	1/7/98 buy	1032.30
	4/8/98 exitlong	1158.60
	profit		 126.30

Test 3: real data - roll at the normal time (Wednesday close)

	1/7/98	buy DSP9H	 974.00
	3/11/98	exitlong DSP9H	1068.60
	3/11/98	buy DSP9M	1080.10
	4/8/98	exitlong DSP9M	1111.80
	profit			 126.30

Test 4: real data - roll 1 month early to avoid volatility

	1/7/98	buy DSP9H	 974.00
	2/11/98	exitlong DSP9H	1023.70
	2/11/98	buy DSP9M	1034.40
	4/8/98	exitlong DSP9M	1111.80
	profit			 127.10

-- 
   Dennis