Hisince long a (T1,T2) FVA (i.e. a forward starting option whoTse strike fixes at future time point T1>T0 & which expires at T2>T1) means being long an Option with Tenor (T0,T2) and short an Option (T0,T1), this trade leaves the holder short the T1 vol & long T2 vol.My question is: does anyone know the formulae to calculate the relevant vegas at T1 & T2 under lognormal assumptions, i.e. NOT the net Vega of the FVA wrt to the fwd vol (T1,T2), but rather the individual bucket vegas = sensitivies to the outright vols (T0:T1) & (T0:T2)Many thanks!

How about flexing each of the vol inputs in your model and then calc the usual v(vol+dvol)-v(vol-dvol)/2dvol/100 (eg London(2005)p143)?

TV = sqrt[ (t2*sigma2^2-t1*sigma1^2)/(t2-t1) ], where t1 and t2 are the times to the fixing of the FVA and the expiration of the FVA respectively, and sigma1 and sigma2 are the market implied volatilities of the atm dns or atmf options of maturities t1 and t2 respectively in years... whether the vols refer to dns or atmf depend on what the fva eventually settles into.. in G10 this usually means dns, in EM (e.g. latams) this usually means atmfsay you buy this FVA in V net vega notional, you can work out the following formulas by differentiating TV wrt sigma1 and sigma2, finding the ratio of one to the other and noting that vega1+vega2=V obviously, then solving for each vega bucket (vega1 being the vega in bucket of tenor t1, vega2 for t2)vega1 = -V / [(t2*sigma2)/(t1*sigma1) - 1]vega2 = V * [(t2*sigma2)/(t1*sigma1)] / [(t2*sigma2)/(t1*sigma1) - 1]sorry dont know how to type in latex but that should sort you out.. enjoy..