Hi,Is it normal that implied volatilities are so little correlated? Is there an explanation for that? I looked at the correlation of 2 stocks, and the correlation usually doesn't exceed 15 or 20% (even when the stock have a 80% or higher correlation). I computed the correlation of the log returns of implied vols on a daily basis over 200 trade days, just as I did for the stocks.Best regards,Jungix

It'd help to know a bit about your study. What time period did you use? Did you use the same strike for each stock throughout the study, or pick each day's 50 delta call, for example? Did you take time into account? If you start with 90 day options and do a 90 day study, you have a quite different animal at the end than you started out with.

the vol smirk guarantees a negative correlation of index implied volatility with price level. so on average, for index options, but for individual stocks... see dealer inventory?

QuoteOriginally posted by: amygdalaIt'd help to know a bit about your study. What time period did you use? Did you use the same strike for each stock throughout the study, or pick each day's 50 delta call, for example? Did you take time into account? If you start with 90 day options and do a 90 day study, you have a quite different animal at the end than you started out with.I used a 200 trading day period to compute the correlation. I used the 3month ATM Call Implied Volatility, updated each day, so there is no problem of getting closer to the expiration date as the maturity is updated, and so is the strike to be ATM. I then have the data for each stock, and I computed the correlation between the 2 log returns in vol over this period.I know that the correlation between one stock and its vol is negative, that's what I found for every stock I tried. When I conpute the correlation between the volatility values (not returns), I had a higher correlation of around 0.5. But I don't know if it is relevant, since the correlation between the stock values is higher too.

A 120 day period would almost certainly encompass an earnings cycle, a time the non-correlated information is having the maximum impact on volatilties. If the two earnings dates aren't near each other in your two stocks, then there would be very little time for the correlated information to stand out from the noise of the news-cycle.

That's true, but earnings and dividends also affect the value of the stocks, and it doesn't hamper their log returns of being highly correlated. Maybe the impact is greater on volatility but I still don't understand why the correlation is so low overall, since I thought there would be similar trends, even if earnings might create some noise.I forgot to precise: I consider stocks that belong to the same sectors (Banks, Oil&Gas, Computers etc) to ensure a higher correlation.

Traders typically build up volatilty well before the actually earnings are announced. The actual move is only one day, but volatilty can trade up and down in anticipation of that move. How many pairs of stocks have you looked at? The 15% you gave _does_ seem low, although we have been in a period of low-correlation for awhile now, at least as defined by all the dispersion traders.

Yes but how dispersion traders define it? I wanted to know if I compute my correlation the same way as they do. I've looked at a few dozens pairs. Sometimes the correlation was even negative, which seemed really surprising to me. I know that the correlation of the stocks doesn't imply the one of the implied vols: I've read an article about the GARCH model and for a correlation in returns of 90% and the 2 GARCH parameters equal to 0.9 each, it claimed that the correlation in volatilities was 12% (and lower with lower parameters). But I don't know much about this model and what these parameters look like in practice.

Jungix;A few things:1. In your first post, you say you got a correl in vol of 15%. In your second post its 50%. Which is it?2. A 15% correl in implied vol is indeed low if returns have an 80% correl. Its either option market specific, or the return correl is spurious. Which stocks are they btw?3. Technically, IF you can fit a garch model, the parameters will reveal whats going on. Correl of vols is a function of the return correl as well as the GARCH persistence parameter. At one extreme if you have constant vols on both stocks (ie: GARCH params =1) then even if returns are totally independant, you'll have high correl in vols. At the other extreme, returns can be correlated close to 1, but if the GARCH params are close to zero (ie pure noise persistence), you'll get no correl in vols.4. If you dont get GARCH param of at least around 0.85 dont use the pair. 5. Doing this on individual stocks is hazadous.6. How you measure correl of impl vol doesnt really matter...what you want anyway is strenghth of mean reversion so measure that directly.

Thanks for your help Vesel.Sorry I wasn't clear enough.1-2. My correl in the log returns of vol is 15% and the correl in vol is 50%. Likewise, the correl in log returns of stock is 80% and the correl in stock is 90% (I just tried this for one stock, TGT vs DELL I think). But I tried for several pairs of stocks and the correl in vol was always low.3-4. I don't know a lot about GARCH models actually and wouldn't know how to compute the parameters (but I thought I didn't necessarily need them).6. You're right, I would like strength of mean reversion. I read that cointegration can be more efficient than correlation (for stocks). Is it the same for vols? Do you have a paper explaining how to use cointegration? Or were you thinking of something else?

Am I wrong in thinking that you shouldn't have to do the log-returns of volatility as it is a stationary series? Of course, stock prices require log returns. Edit: Actually, I guess if you did the Augmented Dickey Fuller to test for unit root (and cointegration) then you then wouldn't have to do the log returns.