As I understand it, a FVA is a swap on future implied at-the-money volatility, which is hedged by a forward starting ATM option / straddle.

I believe the idea behind this is that the future ATM IV is a proxy for expected future realised volatility. But the ATM IV, spot or future, is not a good proxy for expected realised volatility if there is substantial correlation between the underlying and the volatility.

A forward start volatility swap is really a swap on future realized volatility. In another thread, I wrote that Rolloos & Arslan wrote an interesting paper on model-free price approximation of spot starting volswap.

In a very recent working paper (quite condensed) I saw that Rolloos has derived a model-free price approximation for forward starting volswaps as well:

Rolloos - model free forward start volswap

The maths in this latest paper looks correct - but I have not seen numerical tests of the model-free result yet. Anyone tested the latest result by Rolloos, any comments/ideas on it?

As I understand it, a FVA is a swap on future implied at-the-money volatility, which is hedged by a forward starting ATM option / straddle.

I believe the idea behind this is that the future ATM IV is a proxy for expected future realised volatility. But the ATM IV, spot or future, is not a good proxy for expected realised volatility if there is substantial correlation between the underlying and the volatility.

A forward start volatility swap is really a swap on future realized volatility. See for example:

Rolloos - model free forward start volswap