In (value at) risk calculations, we are commonly interested in the risk of changes of the value of our portfolio that are induced by external factors, i.e. thru changes in market prices.
To that end, we usually
fix the invested asset universe and the market environment (e.g. rates / prices / vols) at the onset of our risk calculation and compute a base portfolio value.
We apply a number of shocks, or scenarios, to the market environment and reprice our portfolio using a valuation model. The source of these shocks could be randomness or historically observed shifts or whatever we like.
We might want to consider the change in calendar time as well by rolling our portfolio forward by the calendar length of the market shift, e.g. 1D, 5D or 1M.
Given a sufficient number of repetitions of 2/3, we deduct the simulated prices from the initial price to get a profit-and-loss (PnL) distribution.
We compute some statistic from the PnL-distribution such as VaR, Expected Shortfall or the like.
We may consider shifts in portfolio composition as well. This is usually called management intervention (at least in banks where I worked) and there you could model hypothetical actions. But most of the time, we do not do that as we are simulating the change in portfolio value between points in time where we rebalance our portfolios, e.g. daily or once a week or so.
HTH?