Calculation window

The calculation window changes the weights a method produces without changing the method itself, because it changes the span of history the method measures over. The app's own note reads: "Historical window used by the active method for volatility, correlation, beta, and similar calculations. Default 12." A short window makes the weights react quickly to recent conditions and shift noticeably from one rebalance to the next; a long window averages across more market regimes and produces steadier weights that respond slowly. Neither is more correct than the other — they answer different questions about the same instruments.

The window is measured backwards from each rebalance date inside the backtest, not from today, so a method using a 12-month window at a rebalance in March 2015 reads 2014–2015 data, not recent data.

Ranking-based is the conditional case: it reads the window when its ranking metric is Price Change, Volatility or Sharpe ratio, and ignores it for Average Momentum and P/E Ratio. When the active method does not read the window, the In-sample field is not displayed at all — there is nothing for it to set.

A method that reads annualised volatility over the window accepts only 1, 2, 3, 6, 12, 18 or 24 months; every other method takes any whole number of months from 1 upward. The seven-value case covers:

• Inverse volatility with the risk measure Annualized volatility, in a strategy and in a Combined;
• Min correlation, always;
• Ranking-based with the ranking metric Volatility, in a strategy and in a Combined;
• Beta neutral ranked by Volatility, for its Ranking window.

For those methods the field is a menu of the seven values rather than a number you type. A strategy saved earlier with another value still opens and still saves — the menu shows that value as the current, greyed-out item — but its backtest does not start: the strategy alert "The strategy does not run with this in-sample period" blocks it until you pick one of the seven. On a strategy holding a single picked instrument, Inverse volatility and Min correlation have nothing to weigh and never read the window, so the alert stays silent there.

A strategy holds five ETFs and uses Inverse volatility, which gives each holding a weight inversely proportional to its risk. With the calculation window at 6 months, the weights are set from the last six months of returns: an ETF that was turbulent through that half-year but calm before it measures as high-risk and receives a small weight. Change the window to 24 and that same turbulent half-year is averaged against eighteen quieter months, so its measured volatility falls and its weight rises. Same method, same five ETFs, same rebalance date — a different allocation, purely because the method was shown a different span of history.