Walk-forward evaluation of a municipal futures timed fund switch

What the cash vehicle forces you to test

A long-term municipal bond fund was treated as a constant-maturity fund: it keeps swapping bonds so average maturity stays in a stated long range, and the share price never matures to a fixed par amount. A later sale can therefore return less than the amount invested. An index proxy that times that vehicle has to be evaluated as one mechanical trading system, including the times the system stays out.

The futures stand-in

Municipal bond futures were used as the intermarket driver. Listed futures were described as the first place over-the-counter, hedge, and speculative reactions appear. The fund stance is taken from that related listed contract rather than from the fund's own price.

A continuous daily municipal bond futures series was paired with a fund series adjusted for capital-gains distributions but not for monthly interest payouts.

The two-threshold mechanical rule

The mechanical trading system bought the fund when the municipal bond futures close rose by a set jump-threshold from the prior close, or recovered a set delta-threshold from the short-period futures low. It sold on the mirror conditions measured from the long-period high.

Up and down price-change thresholds were forced to be equal, and a jump-threshold was added. That kept the free parameters few and still let the rule react to a large consensus-shifting close.

Same-session execution and the yearly switch budget

Same-session execution was required. Futures settled early enough that a fund order could still be placed the same afternoon. Next-day fund fills were not accepted as the evaluation standard.

A switch-constraint, the fund-imposed yearly limit on buying and selling shares, made the evaluation favor a low-trade rule set. Alternatives that traded much more often were set aside even when they ranked higher in the in-sample screen.

Fit on one half, then reverse the halves

The sample was split into two roughly equal windows. Delta-threshold and jump-threshold were chosen on one window, with a preference for few trades, then applied unchanged to the other window. The procedure was reversed so each window served once for fitting and once for testing.

Walk-forward analysis here means optimizing entry, exit, and stay-out rules on one window, testing them on the other, then reversing the windows so the same procedure is run in both directions. Agreement between the two reversed out-of-sample windows was not treated as proof that the same parameters would forecast a later stretch of fund prices.