Volatility Model

A more evolved model will include some type of Jump metric in it. Stock prices tend to move around earnings announcement or other event outcomes.

 

opt789

"...conditionally biased" in what way I wonder ?

If this were true then it would imply an edge.

 

Two more quote from papers on the topic:
"The weight of the evidence indicates that implied volatility is not an efficient predictor of intraday volatility but might be an efficient predictor of the volatility of daily futures prices."

"Research has consistently found that Black-Scholes implied volatility (IV) is a conditionally biased—overly volatile—predictor of realized volatility across asset markets. A given change in IV is associated with a larger change in the RV."

Changes in implied vol do give you information about future realized vol but for them to consistently make you money they have to be an unbiased and efficient predictor, which the research has shown they are not. If market makers run out of room to sell then they raise implied vol. It is as simple as order flow imbalance. This change in the implied may or may not lead to a future corresponding change in the underlying on which you can capitalize.

 

Your opinions have merit, however, you imply that certain “random” factors cannot be quantified. In your example you mention accounting scandals and takeover bids. While these events are non-deterministic in nature, general trader sentiment can be modeled by price volume analysis. There do exist neural network models that attempt to forecast M/A activity and bad news via irregular price/volume identification and general trader sentiment. The inner workings of these types of models and their profitability are unknown to my self, likely because if one had such a strategy working, very few would actually hear about it.

I agree with your point about analysis overload, however, you ignore the conditional variance as a useful tool in assessing future risk. Suppose you knew that there was a 10% chance that tomorrow’s volatility would be 5x higher than yesterday with a 80% certainty? This is a conditional variance or likelihood based on historical events that does provide useful information IMO.

 

This is a great point and gets to the heart of the issue. IV is directly based on the current market price, hence the current IV cannot provide statistically viable information other than what the *price* of the underlying asset already provides.

GARCH is a historical function fit of realized vola, but, the model does not provide a "confidence measure" of it's output - a future risk prediction.

It is my belief that a model with a "confidence measure" of its output is a better way to go.

Any thoughts?

Mike