Square-root stock implied volatility ($\Sigma (t,T{)}^2=-\frac{8}{T-t}\log \frac{C(t,T)}{\sqrt{S_t}}$)

For a square-root stock claim priced by $C(t,T)=\mathbb{E}[\sqrt{S_T}\mid {\mathcal{F}}_t]$, the constant-volatility geometric Brownian motion price is $\sqrt{S_t}\exp (-{\sigma }^2(T-t)/8)$. If $t<T$, $S_t>0$ and $0<C(t,T)\le \sqrt{S_t}$, its unique nonnegative inverse volatility is the displayed $\Sigma (t,T)$. For bounded predictable volatility in $dS=S\sigma dW$, bounds $0\le a\le \sigma \le b$ give $a\le \Sigma \le b$, without independence from the Brownian motion. At zero stock price the claim price is also zero and no volatility is identified by this equation; a separate convention is needed.

For bounded predictable $\sigma$ on $[0,T]$, the stochastic exponential $Z_s=\exp (\frac{1}{2}{\int }_0^s{\sigma }_{u}d{W}_u-\frac{1}{8}{\int }_0^s{\sigma }_u^{2}du)$ is a positive martingale by the Novikov condition. Under $dQ/dP=Z_T$, the Bayes formula for conditional expectation givesIndeed, the square-root stock ratio is $(Z_T/Z_t)\exp (-\frac{1}{8}{\int }_t^T{\sigma }_u^{2}du)$. This exact identity gives the bounds on square-root stock implied volatility by bounding integrated variance pathwise. The Girsanov theorem gives $W^Q=W-\frac{1}{2}\int \sigma du$. The measure is an auxiliary pricing identity for this payoff, not a claim that it is an equivalent martingale measure for the original stock.