If you are going to trade VX futures then you really should be trading the curve, not making front month directional bets. That is the best approach to trading VX using FLYS or spreads but that part is the science. The real art is picking the right months, ratios and combinations.
My VIX Futures Journal
- Thread starter TheBigShort
- Start date
Here is an Easter egg for you. If M1-VIX=>150 then sell M1 at t-28ish then sell ES to hedge. I'll let you figure out the ratio and stop on your own.
You are welcome.
Last edited:
Are there people who have been hired because someone thought highly of their journal on Elite Trader? Maybe I'll start a VIX futures journal -- I have researched VIX futures a lot.
uh....unlikely.
I don't know about ET, but Michael Burry (of Big Short fame ironically) was hired after posting his results and thoughts in Silicon Investor message boards.
A few technical suggestions:
Scatterplots:
1) Show a few horizontal grid lines, especially the one through the centroid (or the zero-line if important e.g. if your y-axis is returns).
2) scale your x and y plot range by marginal sigmas or robust sigmas. And center at the joint centroid or median. This makes it easier to visually infer correlation and r^2 by the slope of the regression line.
3) above a threshold, scale the size of your points by influence (diagonal of hat matrix). This makes it easier to tell how much your regression line is skewed by outliers and which ones are the culprits.
ACF plots: always show a PACF plot when you show an ACF plot. Remember that you take your first guess at the AR spec from the PACF, your guess at the MA spec from the ACF.
On your scatterplot of VixPrice ~ Skew, I don't see a "clear negative relation." I only see that your model is almost certainly misspecified, probably in form as well as ommitted variables. More analysis is needed, start with a plot of the residuals. Also usually you'd put the assumed causal variable (Vix level, the comoon belief is that high vix causes low skew, not the other way around) on the X axis.
Scatterplots:
1) Show a few horizontal grid lines, especially the one through the centroid (or the zero-line if important e.g. if your y-axis is returns).
2) scale your x and y plot range by marginal sigmas or robust sigmas. And center at the joint centroid or median. This makes it easier to visually infer correlation and r^2 by the slope of the regression line.
3) above a threshold, scale the size of your points by influence (diagonal of hat matrix). This makes it easier to tell how much your regression line is skewed by outliers and which ones are the culprits.
ACF plots: always show a PACF plot when you show an ACF plot. Remember that you take your first guess at the AR spec from the PACF, your guess at the MA spec from the ACF.
On your scatterplot of VixPrice ~ Skew, I don't see a "clear negative relation." I only see that your model is almost certainly misspecified, probably in form as well as ommitted variables. More analysis is needed, start with a plot of the residuals. Also usually you'd put the assumed causal variable (Vix level, the comoon belief is that high vix causes low skew, not the other way around) on the X axis.
This is excellent advice. I suspect that BS will move in that direction shortly.
Watch out, rm, BS is a quick learner. Pretty soon he'll be the second large pajama vol trader on the forum.rallymode said:
Yes, more than a few. It has usually ended badly.ajensen said:
wow thanks for all the encouragement and ideas everyone!
Currently I am reading as much as I can from https://onlyvix.blogspot.com/search...00-05:00&max-results=4&start=32&by-date=false. If any of you can link to some papers that would be sweet. I found this one particularly useful.
http://www.nuclearphynance.com/User Files/3480/1. MS_QDS_A_Guide_to_VIX_Futures_and_Options (Mar 2011)1.pdf
Lots of questions here.
If the Vix is the variance swap strike, than the vix futures are forward variance swap strike. Usually, when I put on a calendar spread on SPX root-time vega neutral, I am taking a view on forward vol. What am I taking a view on if I put the same calendar on the vix futs? Lets say short 10 MAY and Long 14 June? It's like forward vol of forward vol? BTW is anyone willing to PM me variance swap/forward variance swap data?
Forward vol formula: sqrt((V2^2 * Dte2 - V1^2 * Dte1)/(Dte2 - Dte1))
) . I'll draw up a new one for you kev, and thanks for the reminder in regards to PACF. I will need the right AR() for the initial arima model on spot. What do you think is the best tool for predicting spot? I was thinking GARCH + some fundamental reasoning for starters (ie. if we have GDP numbers coming up in a few days, GARCH will go out the window).
I will PM you about the skew ~ vix relationship, it's definitely not linear and certainly very noisy.
I will be looking at the 2:3 long JUN/OCT calendar tomorrow.
Currently I am reading as much as I can from https://onlyvix.blogspot.com/search...00-05:00&max-results=4&start=32&by-date=false. If any of you can link to some papers that would be sweet. I found this one particularly useful.
http://www.nuclearphynance.com/User Files/3480/1. MS_QDS_A_Guide_to_VIX_Futures_and_Options (Mar 2011)1.pdf
Lots of questions here.
If the Vix is the variance swap strike, than the vix futures are forward variance swap strike. Usually, when I put on a calendar spread on SPX root-time vega neutral, I am taking a view on forward vol. What am I taking a view on if I put the same calendar on the vix futs? Lets say short 10 MAY and Long 14 June? It's like forward vol of forward vol? BTW is anyone willing to PM me variance swap/forward variance swap data?
Hey rally! Thanks for your input. Wont this strategy get creamed in a melt up as vix usually increases with ES? I would much rather stick with the same product so I don't have correlation risk. It would be nice to scale up as quickly as you did and I dont think the strategy you mentioned will be as capital efficient compared to calendars/butterfly's. I have read a good chunk of your posts and it looks like you are long calendars 1:1 during contango and short calendars 1:1 while in backwardation + spot above M1. Is this still a core part of your strategy? Any reason for not doing the calendars ratio root time? Do you try to predict spot or are you solely looking at the relationship of the curve given spots current price?
Below is the "forward vol" between each month of the current term structure. This is usually how I start my search for Calendar trades on SPX/equities, I am not to sure if it applies for Vix Futs. It seem like the forward vols trade at a pretty hefty premium. Would you (@El OchoCinco ) mind going a little more in-depth on how you find your combos?
Forward vol formula: sqrt((V2^2 * Dte2 - V1^2 * Dte1)/(Dte2 - Dte1))
You are right. After reading @srinir post, I did not have the right input variables. I was using the closest vix future to 30 DTE against spy "30d slope" from ORATS (I was running on 3 hours of sleep
) . I'll draw up a new one for you kev, and thanks for the reminder in regards to PACF. I will need the right AR() for the initial arima model on spot. What do you think is the best tool for predicting spot? I was thinking GARCH + some fundamental reasoning for starters (ie. if we have GDP numbers coming up in a few days, GARCH will go out the window).I will PM you about the skew ~ vix relationship, it's definitely not linear and certainly very noisy.
I will be looking at the 2:3 long JUN/OCT calendar tomorrow.
Last edited:
Trading at a hefty premium might not matter if they do not converge ever. Forward vols and current futures will always have a premium built in because there is no precise vol prediction model. You could lose a lot of money trying to predict vols and many Ph.ds do.
You have a curve and spot VIX which the front of the curve converges to and occasionaly is in contango. The rest of the curve slopes down into front month future but it is not a fixed parabolic curve or linear relationship. I ran tests simply running correlation of each VX month to VIX spot to show how parts of the curve are more correlated than others. Even without that research it is pretty easy to simply see the front part of the curve is more sensitive to vol changes then the back part and if Vols rise or fall the whole curve rises and falls but to different degrees along the curve.
The calendar is in one way shorting a month and going long another for partial hedge. You are looking for the spread to widen. So any edge in this trade is getting into the trade when the spread is relatively or unusually small and the expectation is that it will widen as we slide down the curve with time passing. It is simple in its idea but quite complex in actually doing it right.
You have a curve and spot VIX which the front of the curve converges to and occasionaly is in contango. The rest of the curve slopes down into front month future but it is not a fixed parabolic curve or linear relationship. I ran tests simply running correlation of each VX month to VIX spot to show how parts of the curve are more correlated than others. Even without that research it is pretty easy to simply see the front part of the curve is more sensitive to vol changes then the back part and if Vols rise or fall the whole curve rises and falls but to different degrees along the curve.
The calendar is in one way shorting a month and going long another for partial hedge. You are looking for the spread to widen. So any edge in this trade is getting into the trade when the spread is relatively or unusually small and the expectation is that it will widen as we slide down the curve with time passing. It is simple in its idea but quite complex in actually doing it right.
Flies are super capital intensive. Have you looked at the margins?
IMO, correlation risk is better than delta risk. And sure it gets creamed but if it earned in all possible scenarios it would be an arb. Let us know if you find one in vix futures. Especially with M1 or M2 capacity
Last edited: