Before the Death Cross: A Trend-Filter Discipline for MPF Switching

Track 2 — The Tactical Desk | 29 September 2026

MPF cannot day-trade: every constituent-fund switch executes at an unknown forward price, T+1 or T+2. That single mechanism dictates everything — any trend-following rule must be deliberately blunted before it enters the MPF frame. What follows is a 50/200-day moving-average trend filter rebuilt for unknown-price execution: slow, dull, banded, with at most twelve decisions a year.

The Macro Matrix: Neutral-to-Defensive

The dashboard, as of the 28 September close:

• Technicals: The Hang Seng closed at 24,642, up about 0.5% to end a three-day losing streak, but still 1.3% below its 20-day moving average (around 24,960). The index is down roughly 3.3% over the past 30 days in a 24,256–26,009 range, sits 11.6% below its 52-week high of 27,887, and the 14-day RSI reads 42.2 — neutral, leaning soft. One-month return: -3.0%; one-year: -6.3%.
• Yields: The 10-year Treasury broke above 5.2% (highest since 2007), the 30-year above 5.5% (highest since 2004), and the 2-year jumped about 17 basis points in a single week last week. Term-premium normalisation is now the regime; the discount-rate penalty on equities is not finished.
• Liquidity: The Fed hiked 25bp on 16 September — its first hike since 2023 — and the HKMA base rate moved to 4.25% in lockstep. CME futures price a better-than-70% chance of another hike at the 27–29 October meeting. This week brings August PCE (Wednesday) and payrolls (Friday): a data week, so budget for volatility.
• Region: The KOSPI fell 2.7% in a day to 6,890; the CSI 300 dropped 2.2% to a one-year low as mainland money de-risked ahead of the long holiday; US equities fell across the board on Monday, the S&P 500 down 0.8%.

Matrix verdict: neutral-to-defensive. The trend is weakening but no long-term reversal is confirmed. The point of the discipline is not “sell everything today” — it is to write down, in advance, the trigger conditions for cutting Hong Kong equity exposure. Set the rules before the death cross arrives.

Backtest and Win Rate: Buying Left-Tail Insurance, Not Alpha

The rule itself is two lines, adapted from Meb Faber’s 2007 paper A Quantitative Approach to Tactical Asset Allocation and its 10-month moving-average timing framework, translated to a daily version:

1. If the Hang Seng’s month-end close is above its 200-day moving average, hold risk assets; below it, cut back.
2. A 50-day crossing below the 200-day (death cross) shifts the portfolio to defensive; a golden cross restores the offensive stance.

Faber was strikingly honest about what this buys: “a market timing solution is a risk-reduction technique rather than a return enhancing one. The empirical results are equity-like returns with bond-like volatility and drawdown.” AQR’s century-long study of trend-following (data back to 1880) confirms the effect persists across equities, bonds, commodities and currencies; independent replications find similar results for any lookback between roughly 100 and 250 days — the rule does not depend on a magic number.

The Hang Seng’s own textbook case is 2021–2023: after topping out and the moving-average structure flipping bearish, the index fell 14.1%, 15.5% and 13.8% in three consecutive years — a cumulative 37.4%. Against the average MPF balance of HK$343,242, one three-year bear market vaporises roughly HK$128,000; a filter that truncates even half of that left tail is worth about HK$64,000 — nearly a fifth of the average balance.

But the cost must be written down alongside: whipsaw. Moving-average rules get chewed up in sideways markets, and in MPF every whipsaw executes as a blind buy and sell at unknown prices. A BCT programme once revealed a member who switched funds 110 times in six months; Morningstar’s Mind the Gap 2026 quantified the behaviour tax — investors lose 1.2 percentage points a year to timing, which on HK$5,000 a month over thirty years compounds to HK$1.27 million. Hence the filter must be slow: the 200-day line, not the 20-day; month-end confirmation, not daily signal-chasing. Dullness is a survival feature under forward pricing.

The Switching Blueprint: Three Gears, Exact Percentages

Every allocation below maps to real MPF constituent-fund classes (Hong Kong equity / HSI tracker funds, North American equity, Asia-Pacific equity, bond funds, the MPF Conservative Fund, and the DIS Core Accumulation Fund):

Offensive (month-end close above the 200-day MA, 50-day above the 200-day)
Hong Kong equity 50% | North American equity 25% | Asia-Pacific equity 15% | Bond funds 10%

Neutral (signals conflict, or price inside a ±2% band around the moving average)
DIS Core Accumulation Fund 60% | Bond funds 20% | Conservative Fund 20%

Defensive (month-end close below the 200-day MA, or a confirmed death cross)
Hong Kong equity 15% | North American equity 15% | Bond funds 40% | Conservative Fund 30%

Two footnotes. First, North American equity keeps a 15% weight even in defensive mode — it is the portfolio’s only genuine diversifier (dollar assets, low correlation with Hong Kong equities); defensive does not mean all cash, and with the prescribed savings rate at 0.001% a month, all-cash is surrender to inflation. Second, members over 50 already on DIS auto-de-risking should not stack this tactical layer on top — two de-risking mechanisms equal over-trading, which defeats the discipline’s purpose.

The Unknown-Price Shield: Five Rules

1. Monthly review only: read the signal after the close on the last trading day of the month. Twelve decisions a year, maximum. Daily noise is not your business.
2. A ±2% tolerance band: a moving-average cross of less than 2% counts as a false breakout. No action.
3. Two-month confirmation: any gear change requires the same signal at two consecutive month-ends. MPF switches settle at T+2 unknown prices; the cost of a false signal is a blind round-trip, so confirmation is a necessary evil.
4. Blackout windows: no switching in rate-decision weeks (27–29 October), around long holidays, or in major data-release weeks. Unknown pricing plus event risk has negative expected value.
5. One clean switch: staggering a switch means multiple blind executions; intra-scheme constituent-fund switches are generally fee-free (MPF General Regulation s.34(1)), so execute the target percentages in a single instruction and minimise time out of the market and cash drag.

A final restatement of Faber’s framing: this discipline does not buy excess return; it buys left-tail truncation insurance. HK$5,000 a month for thirty years compounds to about HK$6.10 million at 7% but only HK$4.16 million at 5% — two percentage points are worth HK$1.94 million. The trend filter earns precisely that kind of money: the money you do not lose. The 200-day line is unbroken — set the rules now, so that when the death cross comes, all you need is execution, not courage.