Has anyone compared the results of actual testing backtest screen results with buying the stocks in a virtual trading account for say 3 months and compared the results to see how much difference there is between the two?
I have not compared RW to actual results, but in my 6 January post on "Accounting for real-world factors in backtesting", I included a VBA subroutine which attempts to check RW's stated backtesting results for a given strategy. Even assuming zero transaction costs and zero price slippage, I invariably find smallish differences each period between the calculated portfolio return and that supplied by RW; those differences are always in the direction of lower returns than those stated by RW. Moreover, the annualised returns over long periods which I calculate are invariably much lower than those stated by RW.
I have as yet been unable to discover why such discrepancies arise. In some cases, the real-world requirement that one buy integer units of stocks can lead to odd results when the stock price is very high. And in other cases, RW disturbingly assigns a negative selling price to a stock which goes "dead"; I assign a zero price in this case, but this may also be a source of discrepancy.