Forward-Adjusting Live Futures Data for Pair-Spread Models

Summary

The document raises a data-continuity question for a pair-trading model using futures contracts. It says the training series for each contract has been forward-adjusted and asks whether live prices should receive the same adjustment when the model is tested. The example describes a historical adjustment to one leg and considers carrying its adjustment amount forward to a later raw price.

This is useful as a statement of a common continuous-futures alignment problem, but it contains no answer, method, or performance evidence. It does not specify how contract rolls are chosen, how adjustment factors are maintained across roll dates, or whether the model uses price levels or returns. Those details matter: adjustment conventions can alter historical spread levels, while live observations must be made consistent with the representation used to fit the model. The example alone does not establish that adding one past adjustment to every subsequent live price is correct.

Key ideas

• The document asks how to keep live futures observations consistent with forward-adjusted training data.
• Its example considers carrying a historical adjustment from one contract leg into live prices.
• The adjustment convention can affect the price levels used to calculate a pair spread.
• The text does not provide a rule for contract rolls or demonstrate that the proposed live adjustment is valid.

Tags

# For futures contracts, do we need to do price adjustment during live testing?


# For futures contracts, do we need to do price adjustment during live testing?












I am trying to do pair trading on a pair of future contracts, e.g. CME gold and silver.

During the training of my trading model, I do forward adjustment on the pair of future contracts. Let the forward adjustment of a future contract be delta.

My question is, when I test my trading model live, do I need to do forward adjustment of the live data?

I think I need to, for example, let us assume I use 100 days of training data. Maybe on the 90th day, there is a forward adjustment of cme gold, which the raw price on the 90th day is 80, and the forward adjusted price on the 90th day is 100 (which means the delta is +20). Note that there is no forward adjustment of cme silver on 90th day. On the first day of the live testing, the raw price may be 70, so I need to add delta=20 to get the forward adjusted price of 90.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.