adam777 

Hi SMCJB I've followed your posts for quite a while, and with the rest of my study I'm also slowly beginning to understand and see the opportunities that forward curves present. Thank you! I'm wondering where I can find out: 1. how to construct these butterfly "scatter plots" in excel (I understand constructing basic scatter plots, but not these butterfly ones of flys against flys) 2. how to interpret them, 3. and sourcing the data? (not sure if it's standard data from the DDE link from a charting package such as CTS T4?) I've read through CurveAdvisor where it talks about the plots, but not how to construct them. All I need is the name of a textbook, a paper or author, or a hint where to begin looking would greatly help. I'm happy to put in the study to work this thing out. Thank you Adam

Thanks @CobblersAwls for the reply in my absence.

2. how to interpret them
They can be interpreted in many different ways. For example the chart you are linking shows the 18-30-42 fly (which is 12 month fly) versus the 18-42 spread. The 18-42 spread is a measure of curve steepness, but a curve's shape isn't linear, it's normally* logarithmic (for contango and -logarithmic for backwardation). As such this chart is showing that the butterfly was a lot cheaper than it had been historically given the current value of 18-42

* The forward curve is currently a rare M shape.

3. and sourcing the data?
I use Morningstar's Commodity Data Excel Add-In. Not cheap but makes analysis like this very easy.

All I need is the name of a textbook, a paper or author, or a hint
All done in excel. If you plot the forward curve/spread curve/fly curve etc, you'll quickly see some quirks in the shape of the curve. Then just dig deeper looking for statistical anomalies. It's a lot of manual work, especially if your data isn't in a friendly format, which is why I have a lot of this automated.

adam777 

Regards scatter plots I understand double fly price vs time, and fly price vs time, but I'm having trouble getting my head around double fly price vs fly price. I'm guessing you would use time and sales data for the double fly and the fly, but I'm a bit stuck on this one.

I just use settlement data. I'd be careful using T&S as it's rare that all the legs will trade at the same time. I used to do analysis of spreads/flys/structures vs time (days to expiry in this case) AND structure vs structure seperately. Now most of my analysis is multidimensional where I'm looking at "structure vs structure vs time". I find 'R' very useful for that.

Thank you!

I read through the thread on the weekend and learnt quite a lot about crude and trading the curve. There's still a lot of gaps in my understanding but thats' ok..

I guess the most important question is: what is roughly the largest draw-down for a singe trade in ticks/cents before you decide to exit? ... or roughly how far bellow the scatter-plot or curve would you exit?

1 month calendars - 0.0655 - 0.065 etc

3 month calendars (Dec as example) - 0.125

6 month calendars (same Dec Z7) - 0.200
6 month calendars (same Dec Z8) - 0.135
6 month calendars (same Dec Z9) - 0.129

12 month calendars (Z7) - 0.337
12 month calendars (Z8) - 0.266
12 month calendars (Z9) - 0.274
12 month calendars (Z0) - 0.325

What do i get from this?
a) 1 month less volatile than 3 month and 3 month less than 6 month etc
b) for the same type of spread (comparing 6 month Dec calendars between 17,18 and 19), farther you go out, its less volatile i.e 2018 less volatile than 2017 (which is this year)

Note: for the 12 month calendars Z0 is higher vol.. something gonna happen then?? he he .. time to sleep

Can u not do multidimensional regression analysis with excel or is it just better in R

So if u have excel sheets automated with data coming in, though u can visualize comparitve curves I would think to truly identity an anamolies u have to do the regression and see the outliers either just structure or add in the time factor as well for each curve I.e. Calendars then flys then doubled flys to then see which type and month u want to trade

Right?


Sent from my iPhone using NexusFi mobile app

adam777 

I guess the most important question is: what is roughly the largest draw-down for a singe trade in ticks/cents before you decide to exit? ... or roughly how far bellow the scatter-plot or curve would you exit?

I don't have a hard rule for that. There's several things I do that are very different than most here. I don't manage individual trades, I manage everything on a portfolio process, and when it comes to mean reversion systems like this, the further you get from the mean (the more the trade goes against you) the more you add to your position. The biggest thing I do to control risk is to make sure I'm trading things that have time to revert to normality, and to avoid positions where fundamentals could make the market do something I was not expecting. Most of my models/portfolio's do have a risk scaling factor though, so by adjusting that factor I can decrease (or increase) my position sizing.

jokertrader 

100 day Std deviations give u an idea of volatility

Std deviation of what? Absolute prices or price changes? (Textbooks will tell you to use log changes but that doesn't work with spreads where prices can and do go negative.)

jokertrader 

What do i get from this? a) 1 month less volatile than 3 month and 3 month less than 6 month etc b) for the same type of spread (comparing 6 month Dec calendars between 17,18 and 19), farther you go out, its less volatile i.e 2018 less volatile than 2017 (which is this year)

While I agree with you, I would offer a word of warning that if your looking at settlements in months 48(ish) and back they can be 'stale' for lack of a better word. As an extreme example the z24/z25 spread settled -45 last night (and has done for several months) despite the quoted market being -78/-55 for several weeks. This is because settlements back there appear to only get updated when a trades happen.

jokertrader 

Can u not do multidimensional regression analysis with excel or is it just better in R

Yes easy to do in Excel as well. Once you have your data source, and your model setup, it's easier to run multiple different scenario's in R, where as in Excel you would have to set your data up every time.

jokertrader 

So if u have excel sheets automated with data coming in, though u can visualize comparitve curves I would think to truly identity an anamolies u have to do the regression and see the outliers either just structure or add in the time factor as well for each curve I.e. Calendars then flys then doubled flys to then see which type and month u want to trade

I analyze historical prices and build models which I then run against real time prices, so I'm not actually running real time regressions.

As we go farther and farther, i think exchange spread volume is very thin to non existent



Reason i did this is to ask a few questions

a) As we create our model do we need to include exchange prices where available and then include synthetic prices after that so we can get more data points.

B) Columns for the spreadsheet:
we know for calendars it will be 2 contracts per year x say 5 years = 10 columns for calendars

For flys would it be
m17/z17-z17/m18
then
z17/m18-m18/z18
then
m18/z18-z18/m19
then
z18/m19-m19/z19

Then we can
a) graph the structure of the calendars vs structure of the flys vs structure of the outrights (leave double flys out for now)
b) Do the same as above but comparing change between today and yesterday
c) Take the last 100 days of data for each column and run a scatter plot for
(i) All the calendars vs time
(ii) All the flys vs time
(iii) All the calendars and flys vs time.. hmm - does this make sense?

any thoughts/corretions?

Not sure when the attachment was taken but liquidity is higher than that.
As I've said before ZM or MZ front 6 spreads have liquidity, and ZZ front 4 or 5. Outside of that it's very illiquid.

I build z curve first. Then build a Z and M curve next.
It's difficult to build, and also impossible to trade at value, a monthly curve past the prompt 24.

Here is the Crude spreads for 21 days (bar close value)

1. First few columns are outrights.
2. Columns labeled 1 through 7 are calendars - to show column 1 and 2 are the same (difference is prob how the software calculates last from exchange symbol vs the spread bid/ask) - curve chart below
3. 12 month flys and curve below

Couple of interesting oil posts in another thread here. Would definitely recommend listening to the podcast referenced.