Data Integrity Expiration of Futures Contracts

The figures provided in Chapter 2 were either cash market charts, such as spot Interbank foreign exchange (Forex) or cash S&P 500 index, or they were futures contracts for a specific delivery month. This was fine for showcasing how specific technical indicators can be transformed into trading systems, but to generate 10 years of backtested results for a particular trading system on a portfolio, we need to address the issue of expiration of futures contracts.

Nearest Futures Charts The traditional method of dealing with expiration of futures contracts is known as linked nearest contract or nearest futures charting. The nearest futures chart is constructed by including the data history of the futures contract closest to expiration. Following the front month contract's expiration, the chart begins displaying the price history of the new nearest futures contract.

The problem with these charts is that there are usually significant dif-

TABLE 3.1 Composition of backtested portfolio.

Asset Class

Asset"

Asset Symbol

Equity Indices

CME E-Mini S&P 500b

ES

Mid/Long-Term Rates

CBOT Treasury notes

TY

Short-Term Rates

CME eurodollars

ED

European Currencies

IMM Swiss francc

SF

Asian Currencies

IMM Japanese yen

JY

Energy

Nymex crude oil

CL

Metals

Comex gold

GC

Grains

CBOT soybeans

S

Meats

CME lean hogs

LH

Food & Fibers

NYBOT cotton

CT

aTo ensure uniformity, all assets shown are day session only.

fcCash S&P 500 Index x 50 was used to simulate CME E-mini S&P futures.

aTo ensure uniformity, all assets shown are day session only.

fcCash S&P 500 Index x 50 was used to simulate CME E-mini S&P futures.

ferences between the expiring contract's final price and the initial price recorded for the new front month contract. This divergence between the two data sets could result in huge price gaps and, more important, for our purposes, false trading signals. For example, by comparing Figures 3.1 and 3.2, if the February lean hogs contract expired today, the nearest futures chart would rise by 332 points, probably triggering false trading signals in most intermediate-term trading systems.

Equalized Continuation Price Series Charts Most high-functionality data providers enable their subscribers to overcome this problem of false trading signals on long-term nearest futures charts by providing equalized continuation or point-based back-adjusted data series charting. With an equalized continuation series chart, the problem of contract rollover is resolved by the trader choosing a specific number of days prior to expiration day as the trigger for rolling the data in the chart back to the older futures contract month's data series.

Returning to the lean hogs contract rollover problem, if in March 2004 we were to backtest a particular trading system for lean hogs using a equalized continuation price series chart with a designated rollover date of January 19, 2004, as of that date our chart would begin to reflect February 2004

FIGURE 3.2 April 2004 CME lean hogs futures. ©2004 CQG, Inc. All rights reserved worldwide.

data plus the 332-point differential between the February and April contracts. This is because on our designated rollover date the prices were:

February 2004 lean hogs = $5,475

April 2004 lean hogs = $5,807

Our continuous chart would add 332 to all February lean hogs data on and prior to the designated contract rollover date.1

Although equalized continuation charts are a tremendous improvement over nearest futures charts for data integrity in system backtesting, they are not without drawbacks. The first and most obvious problem is that the numbers displayed on these charts are derived through an artificial adjustment of prices, and so the price levels shown are worthless in terms of determining horizontal and trend-line support and resistance and retracement level.

Another problem with equalized continuation charting is that the process of deriving equivalent historical prices often leads to data within the series containing prices of zero or negative numbers. This prohibits our use of stop-loss levels based on a percentage of the contract's value at time of entry. Although we could always refer back to the actual historical prices at the time of entry to derive a percentage-based stop-loss level, there is no need to bother as there are a plethora of equally robust mechanisms for stop-loss placement that can be employed instead.

Point Value versus Percentage Changes in Data History A final issue applies not only to equalized continuation charts, but also to all of historical data. This is the problem of point value changes as opposed to percentage value changes. I will use equalized continuation charts to exemplify the issue. Equalized continuation charts merely adjust the price difference between today's data and historical data, as illustrated by the lean hog example. In many instances, if the asset in question has experienced a long-term bull market trend, then the price differences between entry and exit will be dramatically different from the percentage differences.

For example, let us assume that our equalized continuation chart for Nymex natural gas futures shows a long entry price of $1.001 during August 1991 and an exit price of $1.356 for a profit of $3,450.00 per contract (trade profit was $3,550.00 minus $100.00 for slippage and commissions). Although the absolute price difference between entry and exit levels is correct, if we consider this difference in percentage terms based on August 1991 valuations, we can determine that the actual contract was trading at $1.50 and that a price move of $0.355 represents a 23.67% profit. Now compare this same price move based on October 2003 natural gas prices of $6.00 and our 23.67% profit shrinks to a mere 5.92%.

Thomas Stridsman's book on trading systems addresses these issues in great detail and offers solutions regarding this flaw in equalized continuation data histories. Readers who feel that their backtested results will be affected by such limitations are encouraged to adopt his solutions. In other words, if data are based solely on trading a market with a historical trend similar to the natural gas example, then use percentage instead of price changes.2 However, in pursuing this methodology, remember that the exchange can change the point value of its contracts. Blindly applying a percentage change without consideration of this fact (and of how the software vendor handles such changes) can skew results as dramatically as sticking with the originally flawed price change calculations.3

Examples of other instances in which application of percentage as opposed to price changes would be questionable are the foreign exchange and fixed income markets. Because foreign exchange price increases or decreases are totally dependent on the base currency chosen for valuation, the application of percentage changes are subjective and misleading. This is illustrated by the International Money Market (IMM) Japanese yen contract, which was trading around .003400 in December 1976 and .009200 in

November 2003. Based on these price comparisons, we might erroneously assume that greater weighting should be given to trades executed in 1976 since equal price moves would represent a greater percentage change. This is obviously not the case since the IMM valuation is in Japanese yen-U.S. dollar and use of the interbank market valuations of 298 in 1976 and 109 in 2003 (which are expressed in U.S. dollar-Japanese yen terms) would suggest the exact opposite percentage weightings.

Applying percentage as opposed to price changes to the fixed income market implies a less severe but equally flawed assumption regarding the data. This is due to the inverse relationship between price and yield.4 If an assumption is to be made regarding the application of percentage changes to the fixed income markets, it should be that as prices increase, they may represent lower volatility and therefore would entail a reduced percentage weighting vis-à-vis today's data.

Despite the flaws just detailed, in light of the nature and historical trends of the assets contained with my model portfolio, I remain reasonably comfortable with using equalized continuation charts and have chosen to set the rollover date to 20 days prior to expiration of the contract. Nevertheless, in some instances, where the liquidity was adequate and the correlations between the spot and futures market for a specific asset were significantly high enough, I have decided to use the spot market's data history.

Backtested Portfolio Results Another practical limitation in the presentation of historically backtested results on any significant sampling (for intermediate to long-term systems, 10 to 30 years of historical data are considered a statistically significant data sampling) is the problem of estimating worst peak-to-valley equity drawdowns. To accurately calculate the worst peak-to-valley drawdown on a daily basis, we would need to track daily mark to markets on all assets within the portfolio for the entire data history in question. At the time of this writing, most data vendors with system development and backtesting capabilities do not offer backtested results for a portfolio of assets. Consequently, all worst drawdown and maximum consecutive loss numbers shown in the portfolio totals columns in this and the next chapters are derived from profit/loss and win/loss as of trade exit dates.

0 0

Post a comment