This paper defines the "variance swap delta" as the sensitivity of the price of variance to a small change in the underlying asset price. In plain terms, the authors ask how the market price of future volatility moves when the underlying stock or index moves. They study this question in a mathematical way that connects option prices to variance prices.

To do this they rely on Carr–Madan spanning formulas. Those are mathematical relations used in option pricing that express complicated payoffs in terms of a continuum of option prices. The authors use those formulas to translate assumptions about the shape of the implied volatility curve — often called the "volatility smile" — into a statement about how the variance price reacts to the underlying.

A key assumption they study is when the implied volatility smile is a pure function of log moneyness. Moneyness is a way to compare an option's strike price to the current underlying price; log moneyness is that ratio taken on a logarithmic scale. Saying the smile is a pure function of log moneyness means the whole curve shifts in a particular, scale-invariant way as the underlying price moves.

Their main mathematical finding is that, for that class of smile curves, the total delta of the variance swap is zero. In other words, under these modeling assumptions, the price of variance does not change when the underlying price moves. This is notable because it contradicts a common empirical observation: markets often show higher variance when prices fall.

To address that mismatch, the authors propose a simple modification of the smile so that variance responds to underlying moves in a way that matches the empirical pattern. The abstract does not give full details of this modification or its practical testing, so the scope and robustness of the fix are not shown here. The result is specific to the mathematical class of smiles studied; real market smiles may depend on the underlying in more complex ways, and that dependence must be modeled carefully to capture observed variance behavior.