I discussed at length with Len Yates of OV about the issues related to modeling. I shared with him that having a more accurate IV curve fitting may not result in a better model. In my opinion, a better approach is to focus on forecasting the vol skew instead.
Len agreed with the approach and decided to roll back the modeling to what was in v7.67. After that, he may try to implement the new idea. I think that's great news if we can get back the model that worked well for us in the past.
Please see below Len's reply to me in an email. He asked me to inform the community of this decision.
Kevin,
Excellent information. Good illustration, and I think you have put your finger on the problem exactly. I can even begin to see how a tighter curve fit might be harmful, because the curvature of the tightly fitted line fluctuates more. With a sloppier line fit, the line remains more or less straight.
I’ve decided on a course of action. I will (painstakingly) return the volty skew curve fit algorithm back to what it was in 7.67 and publish a new release. Then, once we make sure that customers are ok using that new version, we can take our time trying to figure out how we can possibly model the curvature change phenomenon that you discussed below.
Len
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Hi Len,
Following up on our phone call, the graph below depicts how the slope and shape of the IV curve change when market moves up. This is an old study that I did. I got this data directly from TOS using DDE to Excel.
1. RUT went up from 1194 to 1210. IV naturally dropped. Blue line is RUT @ 1194 and Orange line is RUT @ 1210.
2. However, IV did not drop evenly across the strikes. Upper strikes, ie those strikes higher than ATM, zone "A" in the diagram, IV dropped more than those in lower strikes, ie Zone B.
3. Look at the table, upper strike IVs declined more in percentage term as well as absolute term.
4. So if I have a butterfly with its right long leg sitting on 1250 and the shorts at 1200. The 1250 will suffer a bigger loss than the 1200 from the IV decline. That'll result in a bigger loss in the position than what was predicted by the T+0 line.
5. Essentially, this example shows the slope of IV curve steepens when IV falls. The reverse is true that when IV increases, the slope flattens. In addition, if you look at the part of the blue line that curls upwards (ie in zone A), the zone straightens on the orange line. I believe this phenomenon introduced the most inaccuracies in the IV modeling.
6. Take note that the slope and shape of the IV curve will also change with time. With everything else unchanged, the closer to expiration, the steeper the slope.
So.... my point is if there is a way to model this shifts in the slope and shape of the IV curve more accurately, the entire modeling accuracy should improve significantly. And to achieve that, perhaps we can start by doing a systematic empirical analysis of historical IV data. I did try to do that before, but to get large amount of historical IV data from TOS into excel was too daunting.
There are better examples but this is the first one I found in my PC. Hope this helps.
Len agreed with the approach and decided to roll back the modeling to what was in v7.67. After that, he may try to implement the new idea. I think that's great news if we can get back the model that worked well for us in the past.
Please see below Len's reply to me in an email. He asked me to inform the community of this decision.
Kevin,
Excellent information. Good illustration, and I think you have put your finger on the problem exactly. I can even begin to see how a tighter curve fit might be harmful, because the curvature of the tightly fitted line fluctuates more. With a sloppier line fit, the line remains more or less straight.
I’ve decided on a course of action. I will (painstakingly) return the volty skew curve fit algorithm back to what it was in 7.67 and publish a new release. Then, once we make sure that customers are ok using that new version, we can take our time trying to figure out how we can possibly model the curvature change phenomenon that you discussed below.
Len
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Hi Len,
Following up on our phone call, the graph below depicts how the slope and shape of the IV curve change when market moves up. This is an old study that I did. I got this data directly from TOS using DDE to Excel.
1. RUT went up from 1194 to 1210. IV naturally dropped. Blue line is RUT @ 1194 and Orange line is RUT @ 1210.
2. However, IV did not drop evenly across the strikes. Upper strikes, ie those strikes higher than ATM, zone "A" in the diagram, IV dropped more than those in lower strikes, ie Zone B.
3. Look at the table, upper strike IVs declined more in percentage term as well as absolute term.
4. So if I have a butterfly with its right long leg sitting on 1250 and the shorts at 1200. The 1250 will suffer a bigger loss than the 1200 from the IV decline. That'll result in a bigger loss in the position than what was predicted by the T+0 line.
5. Essentially, this example shows the slope of IV curve steepens when IV falls. The reverse is true that when IV increases, the slope flattens. In addition, if you look at the part of the blue line that curls upwards (ie in zone A), the zone straightens on the orange line. I believe this phenomenon introduced the most inaccuracies in the IV modeling.
6. Take note that the slope and shape of the IV curve will also change with time. With everything else unchanged, the closer to expiration, the steeper the slope.
So.... my point is if there is a way to model this shifts in the slope and shape of the IV curve more accurately, the entire modeling accuracy should improve significantly. And to achieve that, perhaps we can start by doing a systematic empirical analysis of historical IV data. I did try to do that before, but to get large amount of historical IV data from TOS into excel was too daunting.
There are better examples but this is the first one I found in my PC. Hope this helps.
