This is exactly my point. If I use 6/9/2011 in NT, then my back-adjusted contract on a daily chart will show 39K contracts on Thursday. This is not a reflection of the real level of trading activity on that day, as closer to 100K contracts traded than 39K. I don't want to look at a historical daily chart and assign some special meaning to that day because it had only 39K contracts traded, when in fact the reality was that there was much more volume than that. If it means I have to put 6/10/2011 as the rollover date, then I will have a more accurate picture of activity.
"For example, if the rollover date is Thursday, June 9, 2011, for the S&P 500 futures contract, the CME Globex session beginning that evening (at 3:30 p.m. Chicago time /CT) will list the Sep
2011 contract for trading and the Jun 2011 contract would no longer be available to trade on CME
On the trading floor, the Sep 2011 contract will become the lead month beginning at 8:30 a.m. on
Thursday, June 9, 2011."
I could continue with 10 more data providers, everybody will roll on Thursday.
The reason to roll before the volume has entirely shifted to the new contract, is the high volatility which maybe observed in the old contract, when larger traders have to roll their posirtions. Admittedly this is less of a problem with cash settled contracts, as arbitrageurs will take care of larger deviations. However, for physical commodities that have to be delivered, you may observe huge spikes from time to time.
The following 3 users say Thank You to Fat Tails for this post:
In the document you provided from the CME earlier, it says that:
This seems to indicate that for globex trading, the Friday trading day beginning on Thursday at 3:30PM CT will be the first day of trading for the new contract; only for floor trading does it say that the new contract becomes the lead month on Thursday morning.
@Fat Tails, just curious -- when you say that a continuous contract is suitable for "price analysis over several years," what exactly does that mean? So, it shows real prices, as they traded at that time right? However, what about trend lines based on continuous contracts? From what I understand, these would not show an accurate trend because the continuous contract does not show high to low price swings relative to each other correctly. For reference of a back-adjusted long-term contract, I've attached the back-adjusted crude weekly contract. The caveat may be of course that as long as enough people look at something and trade off of it, that's all that matters, and inaccurate high to low price swings really don't matter, as people are the only thing that makes a trend line "valid" or not.
A continuous contract does not show real prices traded at real-time. It is an artificial construct.
They only way to show real prices traded over real time, is to use single contracts and switch on rollover days. Although this does show real prices, you will get those ugly gaps on rollover days.
If you use a mergebackadjusted contract, this will close the gaps. But in order to do so, the old contracts are shifted vertically. This means that the absolute value of price levels of older contracts is false. It can even move into negative territory. However, the relative values are correct and rollover cost is accounted for. So you can use this type of contract for backtesting. You can apply fib ratios. Only the absolute levels of support and resistance over a longer time period will be false.
The continuous contract does not shift the old contract vertically, but builds an artificial data series from splicing several single month contracts together. This is done by the data provider, and there are hundreds of ways to do that. The advantage of the continuous contract is that absolute price levels over longer periods (10 years) will remain correct. However, all the details are false. If you run a backtest on a continuous contract, it is simply worthless, as none of the values of the data series reflects actual trade data.
So if you only go back a couple of months - as is doen by intraday traders, it is recommended to use mergebackadjusted contracts only. If you are a trend follower trading daily data, only opening and closing trades every few weeks, continuous contracts maybe the better choice, even if they are slightly inaccurate.
The following 5 users say Thank You to Fat Tails for this post:
So if the relative values are correct with the front month backadjusted contract, then a trend line on this chart should show a more true picture of the actual trend, as the relative high-low swings are correct--versus the continuous contract, which has correct absolute price levels, yet incorrect relative high-low swings. Correct?
Is this because for a long-term trader, the actual price number itself may be more important than the technical support or resistance level on a chart? For example, oil reaching a high of $147/barrel, the real value, may serve as a resistance level because it actually traded at that price once (which corresponds to the $184 level on the backadjusted contract). In other words, $147 on the back-adjusted contract means little, whereas that level may be psychologically hard to break because traders remember that it is the high from a few years ago...?
Just for kicks I've attached a chart of the August backadjusted contract with the continuous non-backadjusted contract from kinetick. The backadjusted contract shows just how little we have moved up, relative to how much we dropped, just over 40%, whereas the continuous contract shows this as a 70% retracement.
Trendlines rely on self-fulfilling prophecy, they only work because traders use them. They can be used on any type of chart.
The CL example is the best you could choose. CL rolls every month and has a permanent contango situation,- the Cushing contango -, which is a side effect of the contract only being deliverable in Cushing. For example the Brent Crude contract traded at IPE London does not show a similar contango. CL is actually the worst choice for long only investors, as they do pay more than a haircut for rolling and lose most of their profits.
The contango situation means that the new front month contract comes at a higher price than the old contract month. The old contract month - and all prior contract months - are therefore backadjusted. In contango this means that the rollover gap is added to the old contracts. The difference between $184 and $147 is the sum of the rollover gaps between mid-2009 and mid-2011, so the average rollover gap was about + $1.50.
For CL neither the merge-backadjusted contract nor the continuous contract is entirely satisfying. But again this is the worst example. I do not know any other instrument that rolls on a monthly base and has a permanent contango or backwardation scenario. For all other instruments the distortion is not as bad. You may check this yourself.
Drawing fib retracements from a virtual $ 184 per barrel is not the best solution. Drawing fib retracements on a continuous chart is technically false. In the end fib lines for CL will not work in the longer run, because everybody draws them in a different fashion. This means that they should only be used on shorter timeframes.
Note the difference with FOREX. There are no rollover offsets and fib retracements work particular well, also because competing methods such as floor pivots are not common, as there is no FOREX floor anywhere in the world.
Last edited by Fat Tails; July 14th, 2011 at 09:03 AM.
The following 2 users say Thank You to Fat Tails for this post: