How do you value vol?
- Thread starter .sigma
- Start date
I cannot tell a lie, I was hoping you’d reply to this thread when I thought of it.
your post is very appreciated.
you answered my question but I more now.
you say things like “yangzhang” is my fav on a daily timeframe”
my whole thing is, how could I get my hands on GARCH vol models or YangZhang models? That’s really what I’m trying to figure out.
Let me ask you this, I use thinkorswim and tastyworks as my broker and trading platforms. Would TBS trade based off the data these platforms provide? I’m talking about volatility umbers? Or not? Instead you’d look at your own models?
There are two general approaches to looking at volatility. Both ways are bad, but still are better than nothing.
You can say "stuff used to move this way and this it will continue moving this way" - that means you want to compare current volatility to history. You can do it in a garden variety of ways, anything from simple recent standard deviation to whatever fancy variation of GARCH you can come up with., The main problem with all of them is that you are driving while looking in the rear view mirror.
Then you can say "this other thing saying that stuff is gonna move so this thing gotta move too". In this case you are trying to understand if the implied volatility is consistent with other assets. You can, if you chose so, use all sorts of cross-sectional analysis to make your decision, from simply looking at the historical realized ratios vs current implied ratios to something machine-learning based. The main problem, predictably, is that you are driving a car while looking out from a different vehicle, potentially on a different highway.
You can say "stuff used to move this way and this it will continue moving this way" - that means you want to compare current volatility to history. You can do it in a garden variety of ways, anything from simple recent standard deviation to whatever fancy variation of GARCH you can come up with., The main problem with all of them is that you are driving while looking in the rear view mirror.
Then you can say "this other thing saying that stuff is gonna move so this thing gotta move too". In this case you are trying to understand if the implied volatility is consistent with other assets. You can, if you chose so, use all sorts of cross-sectional analysis to make your decision, from simply looking at the historical realized ratios vs current implied ratios to something machine-learning based. The main problem, predictably, is that you are driving a car while looking out from a different vehicle, potentially on a different highway.
I still use them (in my case Interactive Brokers) to see the quotes but I do not make buy/sell decisions based on their implied and historical vol metrics. As a small retail trader, there is some good money to be made in special situations on small/mid-cap names in the vol space. However, you still need some type of modeling and a way to get your hands on historical data. In order to do this, you need to know some programming or be really efficient in EXCEL.
Sometimes it's better to ask yourself "where might vol be mispriced" rather than asking "how should I price vol". For example, @Same Lazy Element would always find a few cents of edge over me if we were pricing the same options. So I go where he is not looking.
edit* If you want a model-free approach, think about where there might be natural buyers but no natural sellers (or vice-versa). This ties back to asking yourself "where might vol be mispriced".
Last edited:
You seem to be a vol expert
No worries. Most traders do also.
I love Prairie, they bond for life.
The most important concept of vol is they are noise terms, meaning they grow as square root of time.
@TheBigShort, thanks for your reply in this thread. I have a supplemental question. How often do you create these models for a particular instrument? I mean the models probably change everyday very slightly if you consider even the EOD data. Is it the idea to have them automated so that they run everyday and alert on a set criteria? How many underlying to watch and create similar models? I know the answer to the last question is subjective; still wanted to know your thoughts.