|
PureBytes Links
Trading Reference Links
|
To me it seems you are saying exactly the same thing as I tried to say, buy
for a fixed number of dollars when backtesting. Instead of doing the
calculation by hand, I use the fixed dollar transaction entry as you
mentioned last in your mail, it works, excepot for the 65000 dollar share
limit.
The problems i have occur during that kind of backtesting. I do not scale
bet sizes. But without scaling betsizes, you get to my problems if
backtesting over a long period. Ericsson for example is down to closes of
1.8 SEK 1987, many others are down to 0.3 in 1984.
With your own example Num=2000/close, you should get an overrun in number of
contracts if you do that calculation on Ericsson A 1987, should be a value
of 0.2 dollars at close or something like that.
> -----Original Message-----
> From: Bob Fulks [mailto:bfulks@xxxxxxxxxx]
> Sent: den 8 juli 2001 17:02
> To: Bengtsson, Mats
> Cc: omega list
> Subject: Re: Just a note on a shortcoming in contracts
> accuracy in Tradestatio n
>
>
> I generally trade a fixed number of dollars when backtesting.
> Something like:
>
> Num = 20000 / Close;
> Buy Num shares ....
>
> or
>
> Num = 200000 / (BigPointValue * Close);
> Buy Num contracts ....
>
> When trading a fixed dollar of equity exposure, the ideal
> equity curve is a straight line increasing X dollars per time period.
>
> Trading a fixed number of shares or contracts tends to
> distort the results and biases the system to what worked when
> the prices were high. In the NASDAQ, for example. you would
> get big distortions over the past few years.
>
> Increasing the trade size as profits increase (as with "fixed
> fractional trading") will quickly get you to huge numbers of
> contracts where you will hit the 65000 limit. It seems
> totally unrealistic to me since no one would scale bet sizes
> over a long period.
>
> Bob Fulks
>
>
>
> TS2000i allows a fixed dollar transaction in the Format
> Strategy - Cost dialog but I never have used it.
>
>
>
> At 6:06 PM +0200 7/7/01, Bengtsson, Mats wrote:
>
> >This will not cause a problem for many, but it does for me. When I
> >backtest, I want to be as realistic as possible, and also to include
> >both commission and inflation in the calculation in a
> reasonable way.
> >This is destroyed for me by the number of contracts being handled is
> >based on a old wordint with maximum resolution of 65000 contracts.
> >
> >My solution has been to look att the market of Sweden (where
> I live),
> >the commission rules, and the contract sizes. As an end
> result, I do my
> >backtesting based on always buying for a fixed amount (30000 SEK, or
> >around 3000 dollars), and a commission on that. This means that when
> >backtesting on previous years, number of contracts goes up, and the
> >value of each trade will become independent of if it is a
> recent or a
> >distant trade. Thus all signals are valued just as important
> >independent of when they occurred in time.
> >
> >This is easily done in Tradestation , just to enter a trade based
> >commission and a fixed amount in the contract setup for the
> strategy.
> >Now, this works except for symbols that has been around
> quite a while
> >and have had a good price development. For such symbols, the current
> >value will be around 300-400 SEK but the very old value will
> be around
> >0.3xxxx SEK (splits, emissions, ...).
> >
> >I just learned (never got an error message) that
> tradestation can not
> >handle more than 65000 shares in a contract (at least not in the
> >automatic share calculation). When trying to buy 30000/0.333 shares
> >(90090 shares) it decides to buy 65000 shares. Thus eliminating my
> >ambition to have the same value over all years.
> >
> >This also combines with the decimal discussion we had in codelist a
> >couple of days ago. Internally, Tradestation cuts decimals in
> >calculations (I think I can figure out how they implemented
> that, most
> >likely all internal calculations are performed in ticks.
> Thus a price
> >back in year 1984 might be 1.41414. Converted to ticks using the
> >default it becomes 1.414 and number of contracts decided on deviates
> >from the theoretical. Not by too much, but the difference is
> there. The
> >lower the number, the bigger the chance is for a difference from the
> >theoretical, until suddenly the difference becomes huge due to the
> >above mentioned limitation in number of contracts.
> >
> >Well, I now have to decide, do I prefer my 30000 SEK
> tradesize since it
> >gives me a good resolution on 400 SEK shares (one such
> contract more or
> >less in a 30000 SEK trade amounts to around 1.3% in accuracy in the
> >valuebased contribution). Or do I prefer a tradesize of
> 3000SEK which
> >gives me the possibility to have old contracts valued to the
> same value
> >as current contracts, but at the same time introduces a 13% accuracy
> >problem for rounding of 1 share for 400 SEK shares.
> >
> >Ah well, just wanted to give you the knowledge on a possible
> trap, and
> >another example on how the ridicoulusly low accuracy setting in
> >Tradestation can affect us.
> >
> >Anyone backtesting out there that has suggestions on the solution?
> >
> >--- Mats ---
> >
> >
> >This message contains information that may be privileged or
> >confidential and is the property of the Cap Gemini Ernst &
> Young Group.
> >It is intended only for the person to whom it is addressed.
> If you are
> >not the intended recipient, you are not authorized to read, print,
> >retain, copy, disseminate, distribute, or use this message
> or any part
> >thereof. If you receive this message in error, please notify
> the sender
> >immediately and delete all copies of this message.
>
This message contains information that may be privileged or confidential and is the property of the Cap Gemini Ernst & Young Group. It is intended only for the person to whom it is addressed. If you are not the intended recipient, you are not authorized to read, print, retain, copy, disseminate, distribute, or use this message or any part thereof. If you receive this message in error, please notify the sender immediately and delete all copies of this message.
|