in black and white
Main menu
Home About us Share a book
Biology Business Chemistry Computers Culture Economics Fiction Games Guide History Management Mathematical Medicine Mental Fitnes Physics Psychology Scince Sport Technics

Mechanical trading systems - Weissman R.L.

Weissman R.L. Mechanical trading systems - Wiley publishing , 2005 . - 240 p.
ISBN 0-471-65435-3
Download (direct link): mechanicaltradingsystems2005.pdf
Previous << 1 .. 41 42 43 44 45 46 < 47 > 48 49 50 51 52 53 .. 82 >> Next

The next data integrity issue that must be addressed is the accuracy of data. This issue is an absolute prerequisite for developing any meaningful conclusions regarding the success or failure of a particular trading system both now and in the future, and yet this problem is often neglected and/or assumed away.
For most high-end data vendors covering exchange-traded instruments, the problem commonly known as bad ticks has steadily improved over the years in terms of both severity of occurrences and speed at which these erroneous data prints are fixed. Because the ability of each data vendor to handle these issues varies over time, I want to reiterate that accuracy in this area is an unyielding prerequisite for system developers, and it is the one aspect of system development in which superior quality must override any and all cost concerns.
The other issue regarding bad data pertains to non-exchange-traded instruments. It is no accident that with the exception of the extraordinarily transparent and liquid cash foreign exchange market, all the assets highlighted throughout this book trade on a major exchange. Except for cash treasuries and foreign exchange markets, at the time of this book’s publication, whenever we leave realm of exchange-traded instruments, data integrity diminishes dramatically.
Aside from the lack of transparency of non-exchange-traded instruments, the other reason I avoid discussing them is their lack of liquidity. It is worth repeating that underestimation of slippage due to inferior liquidity can have a dramatic impact on the integrity of our backtested system results. Moreover, these effects are magnified as trading time frames are shortened (as exemplified in Table 5.1).
For professional money managers and others marketing hypothetically backtested results to the investment world, I always advise overestimation of slippage and commissions effects on hypothetical performance results. Investors tend to expect future performance to look like past performance. Because this is rarely the case, we must decide whether we want our realtime rate of return to worst drawdown ratios to outperform or underper-form hypothetically backtested results. If an institutional investor allocates $5 million based on an expectation that drawdowns will not exceed 12 percent, the ability to weather a real-time drawdown of 20 percent is probably slim. By contrast, I have yet to hear of a trader losing institutional investment capital due to better than expected real-time profit to maximum drawdown ratios.
Finally, data integrity issues must account for realistic entry and exit
price levels. Many of these issues were addressed in Chapters 3 and 4. Chapter 3 discussed why use of the following day’s opening price on conditionally triggered intermediate- or long-term trading signals was preferable to that of either closing or intraday prices. Chapter 4 argued in favor of assuming losses instead of profits when trading systems achieved profit targets and stop loss levels on the same day.
Other realistic entry and exit levels issues include accounting for gaps beyond “theoretical” fill prices and filtering out trade executions (also known as fills) at opening price levels during a trading day in which futures contracts remain “locked limit.” (Locked limit is a day in which no trading occurs due to a price shock event such as a surprising government report released after trading hours or overnight occurrence of a natural disaster.)1
Data Integrity: Considerations with Backtested Portfolio Results
Another data integrity issue examined earlier was limitations inherent in backtested portfolio results for long- to intermediate-term trading systems. To reiterate, the problem in analysis of a portfolio of assets (as opposed to a single instrument) is that there is no way of determining the portfolio’s real-time worst peak-to-valley drawdown. As a result, system developers will instead look at the worst peak-to-valley drawdown for long- or intermediate-term portfolios based on trade exit dates.
By definition, such sacrifices of data integrity compromise the accuracy of this most essential measure of performance. Nevertheless, if forced to choose between a moderate degree of uncertainty or fuzziness in estimation of drawdowns for a diversified portfolio of assets or absolute accuracy on a single asset, system developers almost universally embrace portfolio fuzziness as the lesser of these two evils.
The reason for system developers’ preference for a backtested portfolio of diversified assets is simple: Backtested results on a single asset can be very misleading, suggesting a losing system where results on a diversified portfolio would show a viable one or, worse still, suggesting viability when a system should be discarded as unprofitable. (For other benefits, see Chapter 9.)
Backtested Data Series: Quantity, Quality, and Out of Sample
There are no absolute answers as to how much historical data is sufficient to ensure the robustness of backtested results. Instead there are only prudent rules of thumb, such as ensuring that our backtested environments in-
Previous << 1 .. 41 42 43 44 45 46 < 47 > 48 49 50 51 52 53 .. 82 >> Next