Trading System Efficiency

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Post Reply
temp
Contributing Member
Contributing Member
Posts: 8
Joined: Tue Apr 24, 2007 11:53 pm
Location: Boulder CO

Trading System Efficiency

Post by temp »

In order for any trading system to yield a return, there has to be movement in the asset prices themselves; the movement provides the opportunity to profit.

I was thinking that a statistical measure of a trading system's ability to turn those movements into returns might be a good parameter for judging the system's effectiveness.

For instance, if an asset price moved from 100 to 110 back to 100 you have 20 points of movement (or, "path"). If a trading system captured 10 of those points it would be 50% efficient.

The key would be in a rational way of calculating the "path". Obviously tick data would yield vastly longer paths than daily data. Perhaps one ought to think of path in terms of movement rather than time ignoring all movements less than some threshold amount, perhaps a multiple of ATR.

I was wondering if enyone else had any thoughts on this. Are there any statistics built into TradingBlox or that attempt to quantify a system's ability to convert the opportunuties presented by a given asset into returns?
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo »

Robert Pardo had a similar idea. Have a look at this post which contains a link to his book (C 1992) and gives the book's definition of "Perfect Profit" viewtopic.php?p=20295&highlight=pardo+perfect#20295. This represents Pardo's idea of 100% trading efficiency. It assumes you are trading only one system on only one instrument, which was a bad assumption in 1992 (8 years after the Turtles traded 2 systems on 20 futures markets) and an even worse assumption today.

A philosophically inclined systems researcher might wonder aloud, whether the world urgently needs more single-number quantifications of trading system desirability, what c.f. calls "measures of goodness". Once you've got
  • the entire equity curve (linear and log scales)
  • the drawdown curve
  • the histogram of monthly profits
  • the standard deviation of daily, weekly, monthly, and annual returns
  • the worst case drawdown measured on daily, and monthly, returns
  • histograms of trade R-Multiples
  • the plot of Margin/Equity ratio
  • the number of shares/contracts/lots traded per year, round-turn, per million dollars of equity
  • the Sharpe Ratio & Sortino Ratio
  • the MAR Ratio & R-Squared
  • the RAR and Robust Sharpe Ratio and R-Cubed from Way of the Turtle
  • the Ulcer Index and Returns Retracement Ratio (Blox Marketplace)
  • Monte Carlo simulations showing how all the above might possibly vary
do you really need more measurements? Is it possible to have two systems A and B which you find equally desirable & identical goodness, based on all the criteria above, such that you need one more measurement to break the tie? Philosophically speaking, of course.
carmelo3750
Contributing Member
Contributing Member
Posts: 6
Joined: Mon May 12, 2003 4:28 pm
Location: Washington, DC

Re: Trding System Efficiency

Post by carmelo3750 »

Eventhorizon wrote:In order for any trading system to yield a return, there has to be movement in the asset prices themselves; the movement provides the opportunity to profit.
I have been thinking about a related issue recently. If a system can derive profits regularly and systematically shouldn't one's attention be turned to what securitites provide the greatest amount of path.

In other words, there are methods we each employ to find, say, stocks, that are primed for a move. We might operate with all stocks above a certain moving average and then wait for a volume, gap. etc. confirmation to trade the stock. Or we might choose the best (or worst) stocks in the best (or worst) sectors.

But I wonder if trading a system isn't more about finding strong candidates for trading and then applying a preferred trading system.

After all, many stocks may cross a 20 day moving average on any day but your capital constraints force you to winnow all candidates to a few. What mechanisms does one employ? The implications are important to a trading system because the ability to find the strongest candidates would reduce whipsaws, early exits, and transaction fees, etc and reduce portfolio variability.

This question comes to mind often but to date one for which I have never found a satisfactory way of thinking of the issue much less a solution.
stancramer
Roundtable Fellow
Roundtable Fellow
Posts: 60
Joined: Sat Feb 07, 2004 10:34 am
Location: Washington DC

Post by stancramer »

Some people divide up the job into three individual tasks and apply a different sets of rules ("algorithms") to each task.

Task1: For each stock, or forex pair, or commodity on the watch-list, calculate an Index of how well that stock is moving. In the recent past, in the present, and in the forecasted future.

Task2: Separate the stocks into Long Candidates and Short Candidates. Sort the Long Candidates by Index-of-good-movement, and sort the Short Candidates by Index-of-good-movement. Only allow yourself to trade the best 6 (or some other number) of Long Candidates, only allow yourself to trade the best 8 (or some other number) of Short Candidates.

Task3: Apply a rule-based algorithmic trading system to today's list of the best 6 Long Candidates. Apply the same algorithmic trading system to today's list of the best 8 Short Candidates.

This seems to be how Trading Blox works. You get to choose your own algorithm for Task1; whatever idea you can imagine that measures Good Price Movement, you can apply. (For example, Chande's RAVI or Wilder's ADX or (RateOfChange/ATR) as in the Blox manual).

Task2 is just a simple Sort, once you figure out how to distinguish Long Candidates from Short Candidates. For some choices of algorithm in Task1, this is remarkably simple: the biggest (RateOfChange/ATR) numbers are the best Long Candidates and the smallest (RateOfChange/ATR) numbers are the best Short Candidates. With other Task1 algorithms (such as ADX) you will have a bit more work to do.

Task3 is a "mechanical trading system" such as the Stochastics System (or the Bollinger Countertrend System) included in Blox. Or any other algorithm for trade entries and trade exits that springs forth from your creative mind.

This method is not perfect, of course. As presented above it will give your task3 trading system the opportunity to trade "the 6 best looking Longs" and "the 8 best looking Shorts." But what if all items on your watch-list are behaving poorly? Task1 will find the (6+8 = 14) best-looking trade Candidates from the set of poor-performing stocks, and task3 will trade them. You are trading the best candidates from among a pool of bad bets, and you will lose money. Take heart, at least you are losing money more slowly than if you traded the worst candidates.

Trading Blox seems to lump Task1 and Task2 together, and call the combination a "Portfolio Manager" module. Task3 seems to be called an "Entry+Exit" module.
temp
Contributing Member
Contributing Member
Posts: 8
Joined: Tue Apr 24, 2007 11:53 pm
Location: Boulder CO

Post by temp »

Sluggo,

Thanks for the link to that thread. It is interesting (to me at least) that GenX' question is along the lines of my broader thinking.

When testing a system against a portfolio of instruments there will be significant differences in the returns for each instrument. In some cases the instrument may simply not have offered much profit opportunity - it's path was short compared to other instruments. In some cases the system was not efficient at converting a path into profit. Finally, some instruments may present sufficient path, but other characteristics make it hard for any system to harvest profit from the path (e.g. trending but noisy).

Carmelo3750 hits on the need to narrow the universe of candidates down by using some ranking system, and stancramer expands on that by putting instrument selection in context of the overall process.

My question in a broader sense is this: are there some instruments that should be discarded because they do not offer sufficient path or their path cannot be effectively harvested by the system under consideration and would thus be an inefficient use of capital?

Furthermore, can one characterise the long-term behaviour of a given instrument to determine if it is moving into a phase where it's path offers less opportunity for profit (either due the path itself or the interaction with one's system)?

In WOTT (pp 216-217) c.f. touches on this briefly in his discussion of different types of markets. In addition, one of Chuck LeBeau's bulletins raised this issue but it never really got developed. I hope to do some work on these issues for futher discussion.
Post Reply