Look at the Figure 1 of this paper:
https://www.casact.org/sites/default/files/old/01pcas_scheel...
There are broadly 2 regimes that dictate volatility and correlation, normal and shit-hitting-the-fan. Risk parity models skew heavily towards the everyday, when prices tick up / down by small amounts, and diversification exists.
On adverse market wide event, there is (generally) no diversification, and leveraged portfolios in particular can face significant losses.
There is no silver bullet, but portfolio wide value at risk, i.e. what the outcome on the day/week/month on any given day in the last 5 years (or more) had I held these exact same positions is as good a measure as any. The distribution of outcomes being something worth understanding and tuning risk to.
The biggest improvement I'd like to make to mine is to implement some approximated form of risk parity[0]. That is, instead of comparing nominal allocations, to compare weighted risk allocations by asset class. This is useful because (for example) equities will contribute significantly more volatility to your portfolio than, say, fixed income, so to the extent you are trying to capture the diversification benefits of allocating across different risk buckets, you may want to scale your exposure according to volatility[1].
There is a modeling challenge here, of course, because asset classes will never be independent risks, but I'd prefer something directionally indicative rather than econometrically optimal.
[0] https://en.wikipedia.org/wiki/Risk_parity [1] https://www.ipe.com/risk-parity-the-truly-balanced-portfolio...