When backtesting a VaR model, do you compare it against the Actual P&L or Hypothetical P&L? If a model passes backtesting perfectly with Hypothetical P&L but fails frequently with Actual P&L, what went wrong?

在做模型回测时，你会用交易台当天的实际盈亏（Actual P&L）还是假设盈亏（Hypothetical P&L）来和VaR做对比？如果一个模型用Hypothetical回测完美，但用Actual回测频频被突破，问题出在哪？