Backtesting #

Using historical data to test the performance of retirement plans is called backtesting. Backtesting is a strategy used in different industries including finance, meteorology, and oceanography.

Backtesting is a one tool, but not the only tool, that you can use to understand and estimate the performance of a particular withdrawal strategy.

Future Performance #

One of the most important things to keep in mind when running backtesting calculations is that they do not guarantee future performance. Just because a portfolio performed well or poorly historically does not mean that it will perform that same way going forward.

Backtesting simulations do not offer certainty: indeed, no retirement planning can, as we cannot know the future. Backtesting calculations provide a host of other benefits, though, including:

  • comparing different withdrawal strategies
  • understanding the behavior of a particular withdrawal strategy in different market conditions
  • determining the conditions under which a particular retirement plan performs well or poorly
  • building confidence in a retirement plan

It is extremely important to keep this in mind when running calculations in FI Calc to avoid misinterpreting the information that is presented in the calculator.

To be clear: even if a retirement plan does well in FI Calc there is no guarantee that it will work out as successfully if you were to retire with that plan today. A healthy retirement plan will involve testing with multiple calculators and calculations. Do not base your entire retirement off of the results of FI Calc.

Other Notes on Backtesting #

Backtesting is only effective when you have high-quality data to run calculations with. The data set used in FI Calc was compiled by a Nobel Prize-winning economist, Robert Shiller (you can read more about the data set used in FI Calc here).

Another problem with backtesting is called overfitting. Overfitting is when you design an algorithm that follows too closely to the current data set, but is not effective with new data.

One of the ways to avoid overfitting is to test your models in other scenarios not present in the historical data, such as by running what are called Monte Carlo simulations. FI Calc does not currently support Monte Carlo simulations, but it is a feature that we plan to introduce in the future.