Firstly, your portfolio volatility of 0.74% is the variance, as the vol will be 8.6% relative your equity position. This is the Case 2 below. I will try to give you a derivation that you hopefully can find an intuition for.
Your portfolio consists of two assets
- A basket/collection of equity with a market price $S$ USD per unit of equity. Assume you hold $n_s$ such units with a total dollar market value $V_s = n_s \cdot S = +\$20\text{m}$. Define the yearly return by $R_s := \frac{S(t+1\text{Y})}{S(t)}-1$, having stdev $\sigma_s$.
- A basket of futures with price $F$ USD per unit of futures. Assume you hold $n_f$ such units with a total dollar market value of $V_f = n_f \cdot F = -\$18\text{m}$. Define the yearly return by $R_f := \frac{F(t+1\text{Y})}{F(t)}-1$, having stdev $\sigma_s$.
Case 1: net the values $\$20\text{m} - \$18\text{m} = \$2\text{m}$
The portfolio market value (exposure) $V_p$ at time (year) $t$ is
$$V_p \equiv V_p(t) = V_s(t) + V_f(t) = n_s \cdot S(t) + n_f \cdot F(t) = \$20\text{m} - \$18\text{m} = \$2\text{m},$$
and stochastic at year $t+1$,
$$\begin{align} V_p(t+1) &= n_s \cdot S(t+1) + n_f \cdot F(t+1) \\
&= n_s \cdot S(t)(1+R_s) + n_f \cdot F(t)(1+ R_f) \\
&= V_s(t)(1+R_s) + V_f(t)(1+R_f). \\
&= \$20\text{m} \cdot (1+R_s) -\$18\text{m} \cdot (1+R_f)\end{align}$$
The 1Y stochastic change (PnL) in value of the portfolio is given by
$$ \begin{align} \Delta V_p(t+1) :&= V_p(t+1)-V_p(t) \\
& = V_s(t)R_s + V_f(t)R_f \\
& = \$20\text{m} \cdot R_s -\$18\text{m} \cdot R_f, \end{align}$$
and the 1Y portfolio return
$$ \begin{align} R_p = \frac{\Delta V_p(t+1)}{V_p(t)} \end{align} = \frac{V_s}{V_s+V_f}R_s + \frac{V_f}{V_s+V_f}R_f = w_1R_s + w_2R_f, $$
where $w_1 := V_s/V_p = \$20\text{m}/\$2\text{m} = 10$ and $w_1 + w_2 = 1$.
The stdev (monetary volatility) of the PnL is
$$ \begin{align} \mathbb D[\Delta V_p(t+1)] &= \sqrt{(V_s\sigma_s)^2 + 2V_s V_f\sigma_s \sigma_f \rho + (V_f\sigma_f)^2 } \\
&= \sqrt{(20 \cdot 0.15)^2 + 2 \cdot 20\cdot (-18)\cdot 0.15\cdot 0.18\cdot 0.85 + (-18\cdot 0.18)^2} \\
&= \$1.7244\text{m} .\end{align}$$
The stdev (volatility) of the Portfolio return is
$$ \begin{align} \mathbb D[R_p] &= \sqrt{(w_s\sigma_s)^2 + 2w_s w_f\sigma_s \sigma_f \rho + (w_f\sigma_f)^2 } \\
&= \sqrt{(10 \cdot 0.15)^2 + 2 \cdot 10\cdot (-9)\cdot 0.15\cdot 0.18\cdot 0.85 + (-9\cdot 0.18)^2} \\
&= 86.22 \% .\end{align}$$
Finally,
$$ \begin{align} \text{Value-at-Risk}_{5\%} &= \text{portfolio volatility} \cdot \text{Z} \cdot \text{Exposure} \\
&= 86.22\% \cdot \text{Z} \cdot \$2\text{m} \\
&= \$1.7244m \cdot \text{Z} \\
& = \$2.8\text{m}. \end{align}$$
Notice that in percentage terms, the portfolio volatility is large, but that is because it is hedged such that the netted exposure is only $\$2m$. the unhedges equity position has a VaR of around $5m. Notice that if you use portfolio returns, and if you market exposure is zero, you will end up dividing by zero. You can then use the "alternative" approach of calculating the stdev as a monetary unit.
Case 2: equity position only $\$20\text{m}$
This is the way you did it.
With $w_1 := V_s/V_s = 1$ and $w_2 = V_f/V_s = -18/20$ so that and $w_1 + w_2 \neq 1$ we get
the stdev (volatility) relative the Equity investment is
$$ \begin{align} \mathbb D[R_p] &= \sqrt{(w_s\sigma_s)^2 + 2w_s w_f\sigma_s \sigma_f \rho + (w_f\sigma_f)^2 } \\
&= \sqrt{(20/20 \cdot 0.15)^2 + 2 \cdot 20/20\cdot (-18/20)\cdot 0.15\cdot 0.18\cdot 0.85 + (-18/20\cdot 0.18)^2} \\
&= 8.62 \% .\end{align}$$
Finally,
$$ \begin{align} \text{Value-at-Risk}_{5\%} &= \text{portfolio volatility} \cdot \text{Z} \cdot \text{Exposure} \\
&= 8.62\% \cdot \text{Z} \cdot \$20\text{m} \\
& = \$2.8\text{m}. \end{align}$$
Case 3: investment amount
We can also calculate returns relative to the cash investment. E.g. we have used $\$20\text{m}$ for the equity positions, and the margin requirement for the futures positions of let's say $20\%$ of $\$18\text{m}$. So the portfolio weights are calculated relative to $\$20\text{m} + 20\% \cdot \$18\text{m} = \$23.6\text{m}$.
Conclusion
As long as you are consistent with how you calculate the portfolio weights and the exposure, you will end up with the same answer. So if you calculate the portfolio by normalising with $\$20\text{m}$, that is also your exposure in calculating VaR.
