Add training workflow, datasets, and runbook

training_data/curated/text/b5ac040fb6bff1d4e944e53412cb5b039367831a2d28d8f822b910addce05423.txt

@@ -0,0 +1,44 @@
+280
+A Complete Guide to the Futures mArket
+series. For example, a 15­year test run for a typical market would require using approximately 60 to  
+90 individual contract price series. Moreover, using the individual contract series requires an algo­
+rithm for determining what action to take at the rollover points. As an example of the type of problem 
+that may be encountered, it is entirely possible for a given system to be long in the old contract and 
+short in the new contract or vice versa. These problems are hardly insurmountable, but they make the 
+use of individual contract series a somewhat unwieldy approach.
+The awkwardness involved in using a multitude of individual contracts is not, however, the main 
+problem. The primary drawback in using individual contract series is that the period of meaningful 
+liquidity in most contracts is very short—much shorter than the already limited contract life spans. 
+T o see the scope of this problem, examine a cross section of futures price charts depicting the price 
+action in the one
+­year period prior to expiration. In many markets, contracts don’t achieve meaning­
+ful liquidity until the final five or six months of trading, and sometimes even less. This problem was 
+illustrated in Chapter 5. The limited time span of liquid trading in individual contracts means that any 
+technical system or method that requires looking back at more than about six months of data—as 
+would be true for a whole spectrum of longer
+­term approaches—cannot be applied to individual 
+contract series. Thus, with the exception of short­term system traders, the use of individual contract 
+series is not a viable alternative. It’s not merely a matter of the approach being difficult but, rather, its 
+being impossible because the necessary data simply do not exist.
+ ■ Nearest Futures
+The problems in using individual contract series as just described has led to the construction of vari­
+ous linked price series. The most common approach is almost universally known as nearest futures. 
+This price series is constructed by taking each individual contract series until its expiration and then 
+continuing with the next contract until its expiration, and so on. This approach may be useful for 
+constructing long
+­term price charts for purposes of chart analysis, but it is worthless for providing a 
+series that can be used in the computer testing of trading systems.
+The problem in using a nearest futures series is that there are price gaps between expiring and new 
+contracts—and quite frequently these gaps can be very substantial. For example, assume the July corn 
+contract expires at $4 and that the next nearest contract (September) closes at $3.50 on the same day. 
+Assume that on the next day September corn moves from $3.50 to $3.62. A nearest futures price series 
+will show the following closing levels on these two successive days: $4, $3.62. In other words, the near­
+est futures contract would imply a 38
+­cent loss on a day on which longs would have enjoyed (or shorts 
+would have suffered) a price gain of 12 cents. This example is by no means artificial. In fact, it would 
+be easy to find a plethora of similarly extreme situations in actual price histories. Moreover, even if the 
+typical distortion at rollover is considerably less extreme, the point is that there is virtually always some 
+distortion, and the cumulative effect of these errors would destroy the validity of any computer test.
+Fortunately, few traders are naive enough to use the nearest futures type of price series for computer 
+testing. The two alternative linked price series described in the next sections have become the approaches 
+employed by most traders wishing to use a single price series for each market in computer testing.