2019issue C016

A trend filter is not ready to compare until portfolio constraints are written down

Readers reconstructing a published trend-filter study asked for the exact position-sizing method used in the backtest. The recorded answers cover cash, capacity, ranking, and shrinking, and they also show which book settings were never written down with certainty.

Reconstructors asked for the exact position-sizing method used in the backtest, not only the trend-filter rule.
Listed settings included a 10000 cash allocation per name, no margin, a 200000 portfolio-size, no-compounding, a 500-name max-open-positions cap, a 100 min-position-value, and a 0.01 per-share commission.
The author confirmed that position-score ranking and position-shrinking were unused, but could not recall every tester setting, including round-lot size.
One reconstructor treated matching the published output as a code-and-settings check before adding extra rules.

Entries in this reading1 entry

Current trackTrend filter137 readings

The missing sizing question

A trend-filter is a quantitative rule that classifies ordered price, volume, or breadth observations over a stated lookback into a directional or participation state. Readers reconstructing a published trend-filter study asked for the exact position-sizing method used in the backtest. Position-size is the cash allocated to one open name in a multi-name backtest.

The author stated that each name was funded with a 10000 cash allocation and that margin was not used.

Listed replication settings

Replication settings listed a 200000 portfolio-size, the starting equity used to scale and constrain the simulated book. The same list used no-compounding, a funding rule that keeps later entries on the original cash budget rather than reinvesting open profits.

The list also set max-open-positions, the hard cap on concurrent holdings the test engine is allowed to carry, at 500 names. Min-position-value, the smallest dollar ticket the test will accept, was set at 100, with a 0.01 per-share commission.

Ranking and shrinking were off

The author confirmed that a position-score ranking function was not applied. Position-score is a ranking field some testers use to decide which signals receive capital when more names qualify than the book can hold.

The author also confirmed that position-shrinking was left disabled. Position-shrinking is a platform option that reduces a new order so it can still open when residual cash is below the target size.

What remained uncertain

Round-lot size was not recalled with certainty by the author, while one reader inferred a lot of 10. A round-lot is the share multiple used to snap order quantities before a fill is booked. The author could not reconstruct every tester setting from memory because the original run was no longer recent.

One reconstructor treated matching the published output as a code-and-settings check before adding extra rules.

Educational research material, not investment advice. Historical source context does not establish present-day performance.

30 of 33 in the Trend filter track

20198-8 pp.Next on Trend filterLookback, threshold, and position-capacity for a stiffness trend-filterThe moving-average and the stiffness-period are free inputs that require parameter-retuning when the sampling interval, the instrument's volatility, or prevailing market conditions change.

All readings on this track · 33 readings

By fieldIndicators and filters682 readings By year201975 readings

All 137 readings tagged Trend filter

Also on Trend filter5 readings

1982An odd-length smoother from a cycle cutoff 1987Testing price-volume agreement after percent reversal filters 1988Auditing the forty-month stock-price cycle 1988Opening-range brackets, a two-bar trend filter, and bounded stops 1988Remaining life on a percent-filtered swing chart