Traders' Roundtable

Just wondering how (or even if) people deal with changing transaction cost in backtesting.

A phenomenon that i've seen in backtesting (particularly when trading at higher frequencies) is that applying todays transaction costs to 10+ years of data leads to a 'flattening out' of the end of the equity curve in many cases (i.e. the closer you get to today, the flatter it gets). I assume that part (all?) of this has to do with the strategy taking advantage of today's costs in a historical environment where they probably weren't realizable.

Any idea what reasonable cost assumptions (commission etc.) would have been in the 70's, 80's ad 90's for futures?

What i've been doing is just setting commission costs arbitrarily high for the whole simulation ($100 / contract) and assuming that if i can survive at that level for a prolonged period, i should be okay.

I've also looked at my 'cost to break even' to see what cost levels would kill me in the long run... seems like a reasonably way to define an upper limit, but i cant really translate it into a more accurate estimation of what real costs would have been.

any thoughts on this?

b-cat wrote:I assume that part (all?) of this has to do with the strategy taking advantage of today's costs in a historical environment where they probably weren't realizable.

What does your curve look like sans costs?

Stanley Kroll's book The Professional Commodity Trader says that in Feb 1974, "Full Commission" on Bellies and Silver was $45, and on Soybeans was $30. (page 91). Whereas Floor Brokerage plus clearing fees were $3.50.

Teweles and Jones's book The Futures Game says that in 1984, "It is not uncommon for large speculators (whose position size requires them to report their holdings to the CFTC) to pay as little as $7 per round turn, as opposed to the $10 to $40 charged to retail traders and hedgers." (page 339). CFTC-reporting Large Specs would include big CTAs such as Richard Dennis and the Turtles.

Bruce Babcock's book The Dow Jones - Irwin Guide to Trading Systems says that in 1989, discount brokers catering to low-volume retail clients charged a commission of $18 per round trip contract. (page 9)

One quick experiment would be to re-optimize your system with commissions set to $45 per contract for the entire test period. See whether the equity curve does or does not flatten out in latter years.

An update on this...

The disclosure docs for many CTAs (example... see pages 17 - 26) contain performance tables that have a column called "Brokerage Commission & Misc Expenses", among various other interesting bits of data.

The commission column shows costs in total dollars / month, so some normalization is required to get a meaningful idea of cost (for my purposes anyway) but the data is there!

It's interesting to see how the relationship between commission and AUM has changed over the last decade