The trading of options on Asian indices like KOSPI, NKY, HSI, etc has become a very competitive game in the
last few years.
Market data acquisition, implying forwards, accurate and robust vol curve/surface fitting, temporal
(aka “smoothing” of microstructure noise, outlier detection, etc), and many other issues have become quite
problems to solve.
Our clients have found our robust and super-fast fitting infrastructure to be extremely helpful in
hard, tedious and time-consuming problem off their plate, so that they can concentrate on where they really
an edge in the trading of these products. Our clients have also found that our sensible and intuitive vol
are very easy to extrapolate into the dark books that exist for maturities beyond 1 or 2 years to expiry.
Before we turn to some examples, a word of caution about the raw data, especially in the wings. We’re using
as inputs to our fitter purely snapshots of bid and ask options prices. The data in the wings often suffer
from “tick size quantization issues” that arise when there is e.g. a bunch of puts = 0.02 x 0.03 for a set
of strikes, followed by a bunch of 0.01 x 0.02 for a couple of smaller strikes. Clearly the best estimate
of the raw option microprices here is not just the mid. Our clients, at least the low-latency ones, use
more sophisticated microprices as input to the fitter, taking into account sizes and higher levels of the
order book, often with a little bit of smoothing over time (our library provides tools to deal with all
this). This resolves many of the tick size quantization issues on the level of the input microprices.
Our fitter is therefore handicapped when receiving the naive mids as inputs, but arguably does an
excellent job even in the wings given this handicap, as we will see.
Now, some examples... we concentrate on the hardest to fit terms in the front months:
C10 vol fit of the first term for the KOSPI. The wiggle on the call side is real (it lasts hours, if
not the whole day) and
has to be fitted for an accurate valuation.
Here both the call and put wings provide illustrations of the tick size quantization issues mentioned
above. For example,
the last two calls shown in the call wing are both C = 0.01 x 0.02 in the data we’re using. The calls
the two strikes just before are 0.02 x 0.03 and 0.05 x 0.06, respectively. Arguably the fitted vol
does a good job given these inputs (and better microprices tend to confirm this: in all cases where we
were able to compare fitted vols and theos from naive and more fancy microprices there is virtually no
difference in the final results).
One could actually use our 12- or 14-parameter curves in this case. Our fitter can fit these high
curves perfectly robustly and the fit in the put wing becomes a bit better below a normalized strike
-6 or so. Whether this is the right thing to do is debatable given the above facts, but, in any case,
fits from the different curves are virtually identical -- and tradeable -- for the 20-odd strikes from
NS = -5 to +2.
C10 vol fit of the first term in the Nikkei.
Note that the zero-bid options in the wings drag down the mid-market prices & mid-market implied volatilities, but Vola's fitter
(robustly) does not drag its implied volatility curve substantially lower in either wing.
C12 vol fit of the first term in the HSI.
The HSI is an interesting example. The data available to us here are of rather questionable quality,
with a lot of fake non-zero
bids (that are really zero-bids), and other quotes, especially for ITMs, that are clearly quite off.
filtering and fitting analytics can nevertheless imply accurate forwards, vols, and arbitrage-free
vol curves/surfaces in a very robust fashion. In this sense, this is a really good example for the
of our analytics.
C12 vol fit of the second term in the HSI.
Total variance plot of the first seven HSI expiries.
Note, these 'total variance' curves, as plotted in this manner, are in sequential order by time-to-expiry and
are non-intersecting — a necessary and sufficient condition for lack of calendar arbitrage.