Evaluating metal seasonal windows across regimes

Season as a core design input

Commodity futures are presented as having firmer supply-and-demand calendars than equities, so timing by season is treated as a core design input rather than an optional overlay. Weather-linked recovery and seasonal demand, for example year-end jewelry demand, are given as reasons metals have calendars, while each metal is said to have its own drivers, so one metal's seasonal window is not assumed to fit another.

A published gold seasonal that looked different

A published gold seasonal with lows in late April and late July to early August, fitted on a 15-year sample, looked different on a 2000 to 2005 window, where a late-April low and a late-July low were followed by strength into year-end.

The last-week rule as one procedure

A last-week rule on weekly bars is encoded as buy when the month matches and day-of-month is at least 23, and sell under the same day rule in the exit month, so the program can locate the final week. Seasonal-trading is a calendar procedure that buys and sells in specified months, and optionally days, so the holding period itself is the signal.

Bounded search over month pairs

When the months are unknown, buy-month and sell-month are treated as system-optimization inputs over that last-week rule, producing a 12-by-12 grid of 144 backtests. Adding day-of-month variables and stepping every other day expands the search to 32,400 runs, or the search can be staged as months first, then days.

Metal-specific windows on one sample

The metals evaluation used weekly series from 4 January 1995 through 30 December 2005 (573 weeks, 4,011 days) and reported metal-specific windows: gold July to December, silver August to March, palladium November to March, platinum September to February, and copper April to February.

Unused dates, fills, and discard rules

After in-sample fitting, the procedure requires a separate out-of-sample window. If that holds, the next gate is simulated execution on live data and fills, because a rule that only works on the design sample is treated as overfit. Walk-forward-analysis re-tests a fitted calendar on unused dates, then under simulated live fills, to see whether the seasonal window survives a new regime.

If repeating the same calendar test across different date windows fails to produce consistent results, the market is treated as having no usable cycle or a cycle that changes too often, and the candidate is discarded.

Confirming technicals when the date conflicts

If a fitted seasonal buy date arrives while the market is making new highs, the calendar is not used alone: trend and support checks decide whether to act, with a break of support ending the trade. Those confirming technicals are independent trend or support checks used when the calendar date arrives in a conflicting market state.