Two rules, and both of them bind on the way in

What the Ghost weighed first

A monthly saving is a small decision made often, which is exactly why the discipline matters. The Ghost does not ask what is hot. It asks three duller questions, in order, and only then spends anything.

Proposed: 6m into ballast, nothing into the bill ladder

5.0m to Government T-bonds. Fixed income is the sleeve furthest below the middle of its band, which decides the sleeve. The curve decides the rung. The 364-day bill has fallen at five consecutive auctions, from about 7.03% in mid July to 6.2852% on 23 September, roughly 74 basis points over ten weeks, while the 35-day rung has risen at three consecutive auctions from about 2.09% to 2.5215%. The short end is getting dearer and the one-year rung cheaper at the same time, which is a flattening from both directions rather than a rally. On the evidence store's last bond prints the government's own curve still pays roughly 523 basis points over inflation at five years and roughly 639 at ten. The proposal would buy ballast where the term premium still exists.

Nothing to Treasury bills, and that is the duration answer. This is the question Chapter 20 declined to answer on a Saturday and sent here instead. September fed the bill ladder at the 364-day rung because that rung paid a real yield where the 91-day did not. It still does, so the reasoning was sound. But the rung has paid less at every auction since, and a sleeve whose reward for a one-year lock-up shrinks each fortnight is one to stop adding to before it is one to regret. The existing TSh 45.6m of bills is not sold, and no rolling ladder is dismantled. The allocation simply stops feeding it while the curve keeps saying the same thing. If the next auction turns the 364-day rung back up, this answer is reversible at no cost, because nothing was liquidated to reach it.

1.0m to iTrust iIncome. A new thought, and a small one on purpose. Before this month the fixed-income sleeve was 86.9% government paper across two instruments, which means the book's entire ballast carried one issuer's credit. A collective income fund diversifies the holder of that risk without leaving the sleeve. At 1m against a sleeve near TSh 198m the proposed arithmetic would barely move, to 86.8%, and saying otherwise would be dishonest: this sets a direction rather than fixing a concentration. On TSh 7m a month you point a book, you do not restructure one. The mandate's own instruction is to prefer adding to earned conviction over reaching for a new name with a small monthly cheque, and this is an existing holding being given a job.

0.5m to cash, and 0.5m to the gold sleeve. Cash sits below the middle of its band, and the half million keeps it there rather than letting it thin further as the book grows. The gold line needs the same treatment for a different reason: it is carried at contributed value, not marked to the metal, so its weight can only fall as the book expands unless savings keep pace. The Bank of Tanzania's August review recorded gold export value 37.4% higher over the year to July 2026, which is the reserve-accumulation backdrop the sleeve was opened against in June. For completeness, and because a reader should not have to find it: the Mining Commission's world reference stood at US$4,291 per troy ounce on 25 September, a fourth consecutive weekly fall, and none of that movement is in this book's value.

And nothing to equities, for a third month running. Repeating a decision is not the same as forgetting to make one, so it is published again, with this month's reasons rather than last month's. The sleeve is the largest in the book and the furthest above the middle of its band. The tape is thin: equity turnover in the week to 25 September fell 48.2% to TZS 27.471bn, a third consecutive weekly fall, and the book marked its highest reading ever inside that same week, which makes the record a reading rather than an achievement. On 29 September both indices eased again, the All Share to 4,650.62 and the domestic index to 10,310.90, while CRDB and NMB together supplied 92.1% of the exchange's equity turnover and foreign sales of TZS 1,514.01m met foreign purchases of TZS 3.62m. The Ghost sold nothing. The proposed growth in other sleeves would take equities from 55.9% to 55.3% without an equity sale. These are projected weights, not a recorded allocation.

Model-book context: how The Ghost reads this terrain → Fixed income · Commodities · Equities · Government securities board

Under the proposed allocation every sleeve would remain inside its mandate band: equities 55.3%, fixed income 31.8%, cash 9.3% and commodities 3.6%. The two collective fund lines together are 9.5% of the book, inside a band of 0 to 15%. No band-breach rationale is required. Nothing was sold. These are the proposal's projected weights, pending ledger reconciliation; no October saving is recorded by this chapter.

The single-name cap: 20%, binding on purchases, and it never forces a sale

Chapter 18 dated this decision to today. Chapter 19 watched the price take the largest allocation back under 20% without anyone deciding anything, and kept the date. Chapter 20 watched it climb to 19.26% and the pair reach 36.52%, all three weeks on price alone, and kept the date again. The date has arrived, so here is the rule and the four questions it was promised to answer.

Should a public model book have a single-name limit at all? Yes, and the honest reason is narrower than it sounds. A concentrated book can be the truthful expression of a small number of researched views, and this one is: five equity holdings, each of which has a published reason to be there. What a concentrated book cannot do is claim that its concentration was chosen, when in fact nobody ever decided it. Over the three weeks of Chapters 18 to 20 the largest holding's weight was set entirely by other people's trading. A limit is how the Ghost takes that decision back, not how it apologises for the holding.

What is the number? Twenty per cent of total book value, per single name, measured on the allocation date. It is set where it is because that is the threshold this ledger's own attention kept landing on for three chapters, and because five names with a 20% ceiling still leaves room for genuine conviction: a holding can be twice the weight of an equal split and still comply. The largest allocation would be 18.55% under the proposed savings, against 18.76% before them; the projected figure remains below the cap. That is the point. A rule that changes this month's behaviour was written in response to this month's prices.

Does it bind on purchases only, or does it force trims? Purchases only. A name at or above 20% becomes ineligible for new savings until it is back under the limit, and no holding is ever sold to satisfy the cap. This is the same answer the duration question received an hour earlier in this note, and for the same reason: new savings are the one thing the Ghost controls completely, while the weight of what it already owns is mostly decided by the market. A cap that forces trims converts other people's price moves into the Ghost's transactions, and Chapters 18, 19 and 20 are three weeks of recorded evidence that a book obeying such a cap would have trimmed, held and trimmed again, paying for each round trip.

And what happens when price drift alone breaches it? Nothing is sold. The breach is reported, each week, in the weekly move, with the weight and the direction, and new savings go elsewhere until it clears. Saying so in advance matters more than the number does: a cap that is silent about drift is a cap that eventually gets quietly reinterpreted the first time it binds inconveniently.

One rule the Ghost considered and declined to write. The pair concentration, the top two holdings together, would be 35.79% under the proposal and has risen three weeks running. It is a genuine reading and it will keep being published every week. It is not becoming a limit, because two ceilings over the same five names start to bind on arithmetic rather than on any thesis, and a second rule written on the same day as the first has no evidence behind it yet. If the pair keeps tightening while the single-name figure stays comfortable, that is the argument for a pair rule, and it will be dated like this one was.

A deposit is not a gain

The public model book dated 2 October 2026 records contributed capital of TSh 521.0m, with July, August and September contributions only. This chapter originally claimed TSh 528.0m as though October had been recorded. That claim is corrected: the proposed TSh 7.0m would take contributed capital to TSh 528.0m only after ledger reconciliation. No October contribution or allocation is added by this correction. The original post-allocation P&L and return comparison is withdrawn as evidence of a recorded event. Contributions are capital, not gains, and the public model book continues to report returns using modified Dietz.

The proposed TSh 6m for fixed income has no recorded October allocation history in this book. The original reasoning was dated 1 October and used holdings marked at the verified Dar es Salaam Stock Exchange close of 30 September 2026. Those observations are retained as the chapter's original vintage, not refreshed to a later mark. Bond and bill lines accrue at indicative yields rather than being marked to auction; the commodities line is carried at contributed value. See the public book and methodology for recorded figures and their dates. October remains pending ledger reconciliation.