• Roll-Your-Own Method 1 (Schwager method). Switch to the newer contract the day after the open interest of the front month contract falls below the succeeding contract for three consecutive trading days. Backward adjust historical prices. (Shift all the prices of all contracts prior to the new one upward or downward by some fixed amount to eliminate any price gap between the two joining contracts on the bar of the switchover.)

• Roll-Your-Own Method 2. Roll to next contract immediately when its volume exceeds the current month.

• Roll-Your-Own Method 3. Roll on a date relative to the beginning of a month. For example, you would roll on the 22nd of the month.

• Roll-Your-Own Method 4. Roll on a fixed number of days from month end. For example, seven days from the months end would be Jan. 24, Feb. 21, and so on.

Whichever method you use, be aware that when producing a very long historical time series of futures prices, back adjustments could create early historical prices near or below zero. This would create big problems for certain indicators, such as price Rate-of-Change. A rate-of-change calculation with zero in the denominator would produce a "Divide by Zero" error fault. That why I have included two additional methods:

• Roll-Your-Own Method 5 (Raw method). Whenever you switch to the next contract, DO NOT backward adjust historical prices. When backtesting, set your trading system to trade as you would in the real world. That is, you sell on the close of one bar, and buy back on the open of the next bar using the newer contract. You will miss out on the overnight price change, but thats the problem with testing on only one time series.

• Roll-Your-Own Method 6 (Log method). Take the natural logarithm of all prices in all the futures contracts intended to be rolled together. When rolling, treat these new prices just as if they were the original prices and apply rollover method: 1, 2, 3, or 4. The negative values that may result are perfectly harmless. When the entire rollover process is completed, take the anti-log of all the prices in this new historical time series. You are now back in the domain of regular prices, but you are guaranteed all prices will be above zero and that all relative price movement (e.g., percent price change or any other indicator based on ratios) is completely accurate, from beginning to end.

Questions to Ask

How are historical futures contracts rolled over? What is the exact formula?

Is there a contract rollover software application compatible with your data format?

If you can verify the rollover software is producing incorrect results, can you get a refund? Will the vendor fix the bug and offer a free upgrade? If so, within what guaranteed time period (e.g., one week)?

Bad Ticks

All day long on the floor of each exchange many "prints" (official records of a trade) are inserted and many are taken out of the official database. Consequently, even though you think you never miss a tick, some of your prices may still be wrong. Your data will include ticks that have been taken out, and your data will be missing ticks that have been inserted. The CME will fax you an official copy of time and sales and you can see how well (or poorly) your ticks match the official ticks. Therefore, it is possible for the market to take out your stop and retreat while your datafeed fails to show the move.

NOTE: You should never assume your trade was executed. Whenever the market gets anywhere close to your stop, it is wise to call your broker and verify if, when, and where it traded

As for accuracy, when traders at the exchange execute a trade at a new price, first they record the trade, and then they yell the price making certain that the price clerks have recorded the new trade. These prices are furiously entered on keyboards (laptops) by the clerks typing in prices as fast as they can. You can see them perched in pulpits above the pit. This manual process creates a large opportunity for bad entries, which propagate down to data vendors, brokers, and traders (you).

Questions to Ask

Does the vendor have bad tick detection and automatic cleanup? What, precisely, is the mechanism or formula applied?

Does the vendor offer automatic bad tick replacement at the end of the trading day? If so, when is it available (how soon after the day session has ended)?

Data Transmission

The second stage in the life cycle of a price quote is its transmission from the vendor to your site.

Generic Issues

Timing your trades is important and, therefore, so is the promptness of your datafeed. In a fast market, S&P prices can move over 200 poinrs per minure. Not all vendors are equipped to handle bursts of market activity. Upgrading datafeed

technology is both expensive and time-consuming. It may take years to complete an equipment upgrade cycle (propose-plan-review-approve-upgrade-test). One year vendor A may be faster, the next year it is vendor B, but only because of decisions made years earlier. Now that the financial world and its needs are changing ever faster, the slow upgrade process is becoming painfully obvious. When the existing technology is not fast enough to handle sudden demands on throughput speed, data may get dropped, and the end user suffers.

Even though much delay is produced by having only one person for each market at an exchange typing in all the trades, the speed of a vendors transmission to your receiver is also a significant factor. As much as a minute and a half difference between a signal at 9,600KB and at 38,400KB was found during fast markets. If you are day trading the S&P, this difference may significantly affect your systems performance.

To improve reliability, some vendors use data buffering, a procedure that stacks information on a queue when data production rates exceeds transmission rates. The hope is that the market will slow down soon enough to permit these buffers to unload. But if the market remains fast for too long, buffers fill and drop data. Curiously, there is a common complaint that European data vendors lose about 20 percent of all ticks on the S&P each day, despite their very fast data transmission rates. Somewhere along the line, buffers are overflowing.

Since it is likely that S&P activity correlates highly with stock activity, those feeds carrying everything (futures, stocks, options, news, etc.) are more likely to have data loss during fast markets than feeds dedicated to only one market. For example, FutureSource transmits quotes on futures and related options and indices. This permits very fast data throughput, especially by satellite, as its transmission is not bogged down with quotes on securities, mutual funds, baseball scores, and so on.

You can test your data feed timeliness simply by comparing tick arrival with another trader (using a different datafeed) over the phone. Another way is to set up a squawk box with the pits and compare what you hear with what you see. A squawk box is a speaker placed by your computer that plays the audio signal from a commentator at an exchange. In addition to comments, you also get to hear floor traders barking in the background. The sudden rise and fall of their overall noise level gives some indication as to what is happening on the floor. Make sure you do this when the market is quiet. In a fast market, the exchanges can be 20 seconds slower (with corresponding delays in the datafeed) than what the squawk box announces.

As an aside, only a small number of vendors create and release their own data: Reuters, Bridge, S&P Comstock, Bloomberg, PC Quote, and Telekurs. Vendors not producing their own data can be regarded as data "repackagers."

Questions to Ask

What is the data rate for transmission? Insist on at least 56KB.

Can you borrow a squawk box for a few days to check out the datas timeliness?