Thoughts on constructing continuous futures contracts? Hi all, How do you construct continuous futures contracts for backtest futures trading systems? There are two components here: (1) Deciding when to roll out the contract and roll into another contract, i.e. jumping from one curve to another curve. (2) Deciding how to adjust for that roll-jump in backtest. My question is with respect to (1). Lets say you have data with lots of years of all kinds of contracts. Some contracts have a life time of 2 yrs and some contracts have a life time of multiple years. One obvious way of using these curves is to stay in front month contract, e.g. rolling every month for crude, etc. But thinking twice made me puzzled: 1. The underlying contracts are actually different product, for some ags futures, the underlying contracts refer to very different crops. 2. For a futures contract that has multiple years of data, jumping each month loses a lot of useful information and potentially trading opportunities. 3. In fact, there is no fixed rule about how to jump among the curves and how should the curves be stiched together. You can jump among the curves almost "arbitrarily" as long as the liquidity permits. Therefore, the problem is about how to organize and utilize these curves effectively... and how to design rules about when to jump and which next curve to jump on to. Any comments? Your thoughts are highly appreciated!
How much money do you make or lose when you roll out of one contract into another contract (excluding commissions & spread)? If you can answer this question correctly then you will know how to deal with the data (for the most part).
That's my question (2), but I don't have problem with that. I just wanted to know how to organize my data and when the curve jumps from one to the other... which is my question (1).
Ok, then, what is your answer to (2)? Alternatively, what is your answer to my question about how much money you make or lose on a roll?
The PDF document above talks about question (2). If you jump from curve x which has a settle price X, and to curve y which has a settle price Y, then you lose (Y-X), or you make (X-Y).
The answer to question 1 depends on how you want to use your data. If backtesting is important, but not real time use, simple collect data for both symbols and when the volume of the new contract goes over the volume of the old one (end of day basis), roll to the new one. You can simply take the closing prices of both to determine how much to adjust the old one. Settlement prices, if available, would be even better. Or determine however way you want to determine the diff. Problem with above approach is, it's hindsight, and we can't trade on hindsight. Say you are running a strategy (actually trading it) on back-adjusted ES data. Usually the volume rolls over on the second Thursday of expiry month. So you can be proactive, and do the switchover after 3:15 PM CST on Wednesday before trading begins on 3:30 PM CST. The benefit of this approach is that you can recalculate your signals, and your strategy can keep trading happily for the next 3 months. What if for some reason the old contract still has more volume than the new on Thursday? What if it does? It's academic and would have no bearing on your strategy. After all, ES has enough volume to accommodate any strategy even if you are one day early (or one day late, for that matter). This approach let's you be proactive, and have your business running. It's what I use. If you study the volume rollover dates, you can anticipate the rollover dates for any contract. For trading, in my books, close enough is good enough.
No, that is incorrect. When you roll from one contract into another contract you make $0 (excluding transaction costs). If you think about how futures contracts make or lose money you will see why. Not to be a d*ck, but have you ever traded a futures contract (or been an assistant for someone who traded futures)? If not (and there is nothing wrong with that) I suggest you open up an IB account and make some small trades on liquid, cash-settled contracts so that you can understand the instruments, and then worry about how to backtest, market microstructure, sugar spreads, etc.
You asked me about the PNL of roll itself. On the roll day, you roll at EOD. The contract you are currently long trades at $100. The contract you are rolling into trades at $200. In order to be out of the current contract, you sell it at $100, and in order to be in the next contract, you buy it at $200, you PNL for this roll is $-100. What's wrong with that?