Quandl Continuous Futures: Adjustment Assumptions and Backtest Bias
Summary
The document asks whether Quandl’s Wiki Continuous Futures series are already adjusted for contract rolls and how to handle them in rolling train-and-test backtests. It distinguishes continuous series from explicitly back-adjusted data: the presence of a continuous contract does not establish that prices were adjusted, and a researcher would need to know both whether adjustment occurred and which method was used.
The answer is brief and offers no roll-detection procedure, adjustment algorithm, or empirical comparison. Its practical lesson is to verify the dataset’s adjustment status and method before interpreting long histories or using them in a backtest. The cited warning about cumulative roll effects motivates that care, but the source does not resolve how to identify roll dates when they are undocumented, nor does it evaluate the proposed window design.
Key ideas
- A continuous futures series should not be assumed to be back-adjusted unless the provider explicitly says so.
- If a series is adjusted, the adjustment algorithm matters for interpreting backtest results.
- The document gives no method for recovering undocumented roll dates or preventing bias.
Tags
Full text
# Using Quandl Continuous Contracts
# Using Quandl Continuous Contracts
I am trying to use Quandl data futures for backtesting some trading scenarios, specifically Wiki Continuous Futures.
Following the documentation, I understand that the data-set contains continuous contracts named by the following convention CHRIS/{EXCHANGE}_{CODE}{NUMBER}. For example for crude oil (CL symbol) I will find various datasets depending on the market and on the back month contract. Eg. CHRIS/CME_CL1 will be the front month of Crude Oil Futures from the CME exchange. Whilst, CHRIS/CME_CL12 will be the 12th back month of the same future, the same market.
Can I assume that given that these are continuous contracts they are already adjusted? But then again, in the documentation there is this statement:
> Of course, one must take care when analyzing and interpreting Continuous Contract data spanning decades, because the impact of multiple Rolls over such long time frames can be quite significant.
LE: I would want to backtest my algorithm splitting the data into windows and applying a non-anchored forward window strategy. Each window should contain approximately 4 years of data: 3 for model training and 1 for testing, therefore unseen data.
My question is: how can I transform the data (ratio adjust algorithms) in order to not introduce any bias in the data? Given that in the documentation there is no indication when the contracts roll.
## Answer by BobL (score 0)
https://quant.stackexchange.com/a/68504
You cannot assume they are back adjusted unless explicitly stated and then you would need to know the algorithm used to adjust so you can take that into account in your back test.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.