Construct a winter seasonal window as one procedure

Treat the window as a procedure

The source constructs a seasonal equity procedure that looks for mid-December new lows and holds into a winter exit rather than buying on a single calendar open. Seasonal trading, in this setting, is a calendar-constrained procedure that specifies when a market may be entered, held, and left, and when the system should stay out.

Editorial: a famous calendar window is a construction problem. Freeze the holding period, then test which dip definition, start date, and exit date survive as one procedure rather than as a story.

Freeze the hold, then define the dip

Cash Dow industrials from early 1920 were used for construction, with data from March 3, 1990 onward reserved before parameter search for later unused testing. That reserved block is an out-of-sample holdout: data set aside before rule search begins so later confirmation is not taken from the same observations used to pick parameters.

Twenty dip definitions were compared by buying one tick below the lowest low of the prior 1 to 20 sessions, holding the original December 15 to February 15 window fixed. The entry channel is that lookback. The simplest one-session dip produced the largest net points in that search and was locked as the entry rule for later date tests. The locked trigger is a mid-December new low: a long entry after mid-December when price prints a new short-horizon low, rather than a market-on-open buy on a fixed calendar day.

Search start and exit dates

System optimization here is a controlled search over rule inputs such as dip lookback, entry calendar day, and exit calendar day while other parts of the procedure stay fixed. With the exit held at February 15, entry calendar days from December 10 through December 20 were compared and December 15 was retained. Exit dates were then searched from January 2 through March 31 to jointly consider net points and the share of winning trades, after which January 7 was selected over February 15.

Check unused data and other samples

A 1920-89 in-sample window and a 1990-99 unused window were reported side by side for average yearly percent change, percent profitable, and average win versus average loss. Walk-forward analysis is that split of historical data into an in-sample construction window and a later unused window used only to check whether the locked rules still behave similarly.

The same locked calendar-and-dip procedure was then applied to short Dow index futures history and to S&P cash and futures samples beginning in 1960 or 1982.

Add a stop after the rules are locked

A 4.5 percent protective stop was examined after the calendar rules were set. Some historical losses still exceeded that limit because early years used close-only data. A protective stop, in this workflow, is a percentage loss limit tested after the calendar rules are set, used to study how large adverse moves interact with close-only historical quotes.