Bizzare EOD Trading Strategy

[Cableguy]

Folks,

Testing on EOD data, my goal is to determain if the next day is a viable trading day, opening the position before the market opens and closing it on close, here's the rules (long only):

Daybefore (eod)
If not (difference between closeprice and highprice is less than or equal to 20%) then

On eod​

IF (difference between closeprice and lowprice is less than or equal to 20%)
Open Position
​

Results on the DOW from 01/01/2002 to yesterday

Trades: 75
Successful: 53 (points: 4506.6192)
Failed: 22 (points: -1581.7500)
Diff: 2924.8692
If maximum high achieved: (6368.5091)

Testing this from Jan05 results in (759.7992) points 1 failure and 10 successful.

Now, on the failed days (2002-2005) all closed lower then they opened however only 5 didn't achieve a high (opened on high).

My thoughts: Open all positions with a trailing stop? Could such a strategy (if it deserves the title) work? Here are the recommended positions produced today (i dont know the outcome as the some markets haven't opened/closed yet):

Indicies:
DJI
SP500
FTSE100

Equities:
ANTO.L
AZN.L
EXL.L
JMAT.L
MRW.L
PSON.L
RSA.L
SDR.L
TATE.L
YELL.L

Forex:
GBPCHF
GBPUSD
EURUSD

Thoughts/Comments?

Cheers

 

[JonnyT]

You cannot trade these using EOD data.

Try again using the futures data....

JonnyT

 

[Cableguy]

Are you refering to all of these (or just the forex)? And why?

Cheers

 

[JonnyT]

Taking the dow for instance.

If you open the position just before the market opens what price have you assumed? The closing price the day before? or the futures price at the precise time?

What time do uou consider the market opens at?

My guess is that you are using yesterdays closing price which is not valid.

In the past I have found superb systems on EOD data that in reality are losers. Some were unbeleivably better than this...

JonnyT

 

[Cableguy]

Hi

The open price is the actual opening price of the next day, according to the EOD data. I did use the closing of the previous day before but eventually realised it was wrong

The backtesting doesn't take in account the market open/closing times, should it? It just reads "if I have a singal, then what were the results for the next day" -:

if (nextdayclose > nextdayopen) then good. This was easier to code as I was testing various amounts of data (forex, lse, nasdaq, nyse) all with different opening/closing times.

Does this qualify as valid backtesting?

Cheers

 

[the blades]

Hi

The open price is the actual opening price of the next day, according to the EOD data. I did use the closing of the previous day before but eventually realised it was wrong

The backtesting doesn't take in account the market open/closing times, should it? It just reads "if I have a singal, then what were the results for the next day" -:

if (nextdayclose > nextdayopen) then good. This was easier to code as I was testing various amounts of data (forex, lse, nasdaq, nyse) all with different opening/closing times.

Does this qualify as valid backtesting?

Cheers

Hi Flubber,

If you want soem EOD futures data for the Dow, let me know.

A few questions;

the difference between the high and the close of the day must be less than 20% - 20% of what?

If so, then you buy on the open of the following day? What's your exit rule - is it simply to close at the end of the same day?

Sorry if I'm missing the obvious here. I believe simple systems like this can work, though the cost of commission and slippage could well wipe out the gains, so make some allowance for this.

As for the validity of your backtesting - why do you suspect that it isn't valid? You have a simple system that isn't over optimised, so if you allow for Costs + slippage, it should be OK - though Jonny is correct re the futures data. That said, it isn't at all certain that it wont work with futures - try it..

Cheers,
UTB

 

[Cableguy]

Hi Flubber,
If you want soem EOD futures data for the Dow, let me know.

That would be great thanks...

the difference between the high and the close of the day must be less than 20% - 20% of what?

20% of the difference between the high and low of that day, for example, the Dow opens at 10,000 reaches a high of 11,000 and a Low of 9,000 then closes at 9,500.

High_and_Low_Difference = high-low (2,000)
Low_and_Close_Difference = close-low (500)

Factor = (Low_and_Close_Difference / High_and_Low_Difference) * 100

Is Factor <= 20? In this instance it was 25, so this wouldn't be a tradeable signal.

The idea is to calculate an area or zone of where the prices are, I hope that makes sense.

If so, then you buy on the open of the following day? What's your exit rule - is it simply to close at the end of the same day?

Yea, the exit rule is the problem, although I've noticed that 90% of the time the underlying instrument does acheive a high during the trading period higher then the open giving opportunity for profit with a trailing stop.

As for the validity of your backtesting - why do you suspect that it isn't valid? You have a simple system that isn't over optimised, so if you allow for Costs + slippage, it should be OK - though Jonny is correct re the futures data. That said, it isn't at all certain that it wont work with futures - try it..

Ok, thanks, I wasn't to sure about the backtesting as to how much data should be tested, both in bear and bull markets, and factors as you have mentioned slippage and commission.

This system does kinda work, the prediction the DOW would go up yesterday was correct, the DOW results always seem very good why would this be?

Cheers

Flubs

 

[mr_cassandra]

back-testing via spreadsheet or program

My advice would be put it over to a machine and let it crunch the numbers to get an objective analysis. It should be able to work under bearish conditions, bullish conditions and a flat market. Make sure you factor in commissions and ask vs bid so the numbers don't look too optimisitic.

You can get a lot of opinions online, somebased on a lot of due diligence by the poster and other are just standard euphemisms they learned about 'that won't work because'.

If your numbers are accurate and you test enough conditions with a program or spreadsheet, and it still looks good then try realtime paper trades and again be accurate.

In the development of my own system, I found that the sterile environment of a 8 year back test was 'not' the same as when I had real money at stake. I also found, as I did my own work,that many of the observations that discussion of my program had engendered were merely quotes from an article of news show the poster had seen. In y own case it became useful to always ask each person with a 'that won't work' about their own research and what numbers they had crunched on their own. My all time favorite was one at yahoo stock chat who said I haven't checked, I won't look at your spreadsheet and I KNOW it won't work. The poster herself had to admit she made little sense.