In the equation $$(5)$$, he is defining the variance of the spread as:

$$\sigma_1^2S_1^2 + \sigma_2^2S_2^2 - 2\sigma_1 \sigma_2 S_1 S_2 \rho$$

whereas I have always seen it defined as:

$$\sigma_1^2 + \sigma_2^2 - 2\sigma_1\sigma_2\rho$$

This is for 2 correlated GBM and the spread is $$S_1 - S_2$$.

What am I missing?

• I think they are moving it into normal returns rather than lognormal, to remove the issue of negative spreads. – will Nov 11 '19 at 22:08

I think the variance of the instantaneous shifts in the spread is meant:

$$V \left[ dX \right]=V \left[ dS_1-dS_2 \right]$$

And the individual variances (in the conditional and local sense) are:

$$V \left[ dS_1 \right]= \sigma_1^2 S_1^2dt$$

$$V \left[ dS_2 \right]= \sigma_2^2 S_2^2dt$$

And the covariance term is, assuming the two Brownians are correlated:

$$C\left[ dS_1 , dS_2\right]=\rho \sigma_1 \sigma_2 S_1 S_2dt$$

Now if plug these into your formula, you get the equation 5.