Nine-tenths of the money, in two names that fell

This window has four verified sessions, not five. 14 October is Mwalimu Nyerere Day, a public holiday; the exchange did not trade and no close exists for it. The Ghost says so first rather than quietly dividing by five, because a missing session distorts every weekly total further down this page, and the distortion is the honest part of the comparison.

From the 9 October close to the 16 October close the All-Share (DSEI) fell from 4,597.48 to 4,572.74, about −0.54%. The domestic Tanzania Share Index (TSI) fell from 10,142.06 to 10,079.13, about −0.62%. Both are third consecutive weekly falls, and both are the smallest of the three: the All-Share has run −1.08%, −0.74%, −0.54%, and the domestic index −1.57%, −1.11%, −0.62%. A sequence of shrinking falls is not a recovery. It is a market running out of sellers more slowly than it ran out of buyers.

Underneath, the sectors stopped agreeing. Banks, Finance & Investment fell from 22,842.38 to 22,471.84, about −1.62%; a third consecutive weekly fall, and unlike the headline indices this one got bigger than the week before it. Across the three weeks from 23,714.23 on 25 September the banking index is down about 5.24%. Meanwhile Industrial & Allied rose from 5,034.51 to 5,122.17, about +1.74%, the strongest sector reading in a month and a reversal of its own fall a week ago; Commercial Services rose from 3,531.40 to 3,557.70, about +0.74%. The spread between best and worst sector index widened to about 336 basis points, from 217 and 265 in the prior two weeks. Four consecutive weeks of narrowing ended here. The market did not fall together; it split, and the half the book is heaviest in is the half that fell.

The sessions ran +0.44%, −0.61%, −0.37%, +0.01%. Breadth across those four sessions was 33 advances against 35 declines, against 33 against 45 over five sessions the week before; close to even, and better than it has been in a month. Friday closed 0.01% higher, which is a rounding error wearing the costume of a recovery.

Friday to Friday, across the 28 counters with a verified close on both dates, 10 rose, 13 fell and 5 were unchanged. So a minority of the board fell and the index fell more than the board did. That is the same arithmetic as last week's dead-even count, one week further on, and this week the mechanism is almost rude in how plainly it shows itself.

Here it is. Across the four sessions, total equity turnover was TZS 12.944bn. CRDB alone accounted for 65.2% of it and NMB for 26.0%: together, 91.2% of every shilling of equity turnover in the week. CRDB led turnover in all four sessions, at 73.7%, 70.5%, 54.0% and 59.7% of the day. And over the week CRDB fell 2.41% (2,900 → 2,830) and NMB fell 1.00% (2,010 → 1,990). The two counters that were this market's liquidity both went down. An index weighted by size cannot do anything else.

This book has written that sentence before. Chapter 12, on 8 August, recorded two banks supplying 85.4% of the week's turnover and called it a concentration to study rather than feed. Ten weeks later the figure is 91.2%. The concentration did not resolve; it deepened. Between 20 and 21 of the 28 counters traded at all in any given session, which is the same fact from the other end: most of the board is priced, and only a little of it is actually liquid.

The individual moves, for the record and without explanation attached. Risers: MUCOBA +7.79% (385 → 415), TCCL +5.26%, KA +4.35%, PAL +3.45%, TPCC +2.65%. Fallers: JATU −14.81% (270 → 230), TOL −8.70% (1,840 → 1,680), NMG −6.15%, NICO −5.04%, DSE −3.74%. Two of those are last week's extremes handed back: MUCOBA fell 14.44% last week and rose 7.79% this week; TOL rose 6.36% last week and fell 8.70% this week. The Ghost offers no reason for any of them, because the DSE listed-company filing register this book reads was last generated on 23 September and has not been regenerated in the 24 days since. On the smallest, thinnest counters, a double-digit move that reverses the following week is more often the bid-offer spread being crossed in one direction and then the other than it is news. Without a filing, that is a hypothesis and the Ghost will not dress it up as a finding.

On the money's direction: foreign sales exceeded foreign purchases in all three sessions the participation record covers, TZS 1.843bn net. That record is missing 13 October, so this is three of the week's four sessions and the weekly total is not comparable to last week's TZS 3.131bn across five. Per covered session it is almost exactly unchanged: about TZS 614m against TZS 626m. Net foreign selling now runs nine consecutive covered sessions, back to 2 October. The headline halved; the rate did not move at all. No motive is inferred; it is the exchange's own participation arithmetic.

And the same caution applies to turnover. TZS 12.944bn against TZS 18.019bn is −28.2%, which would be the fifth weekly fall in six. Per session it is TZS 3.236bn against TZS 3.604bn, about −10.2%. Most of the headline fall is the holiday. The real thinning is a third of what the headline claims, and quoting the headline without the divisor would have been the easiest small dishonesty available this week.

The model book went from TSh 611.22m on 9 October to TSh 611.03m on 16 October; −TSh 0.190m, about −0.03%. It is a third consecutive weekly fall, the first run of three this ledger records, and it is six times smaller than the second and thirty times smaller than the first: −0.90%, −0.19%, −0.03%. Contributed capital did not move, so the whole of it is market P&L.

Holding by holding, against the verified closes: Twiga Cement 5,670 → 5,820, +2.65%, +TSh 0.315m. Tanzania Cigarette 13,380 → 13,510, +0.97%, +TSh 0.579m. Vodacom 1,270 → 1,280, +0.79%, +TSh 0.833m. NMB 2,010 → 1,990, −1.00%, −TSh 1.118m. CRDB 2,900 → 2,830, −2.41%, −TSh 1.189m.

Three of five rose and two fell, and the two that fell are the two bank allocations; which is the banking index's sentence, and the turnover table's sentence, written a third time in the book's own currency. The three risers are the three non-banks, and Industrial & Allied was the week's strongest sector. The book did not diversify its way out of this week by accident; it was built with three names that are not banks, and this is the week that paid for.

The equity sleeve lost TSh 0.580m. The fixed-income and cash sleeves accrued TSh 0.390m. Those add to −TSh 0.190m, the NAV move exactly; there is no third thing. The ballast covered 67.3% of the equity loss, against 23.7% and 6.2% in the two weeks before. Read that sequence carefully, because it is the one number on this page most likely to be misunderstood: the accrual was TSh 0.390m this week against TSh 0.370m in each of the prior two, a difference inside rounding. The coverage ratio tripled because the hole shrank, not because the ballast did anything differently. Income is a near-constant that arrives whatever equities do. A week in which it covers two-thirds of the damage is a week in which the damage was small; it is emphatically not evidence that the book is hedged, and if equities ever fall 5% in a week the same TSh 0.39m will cover about a fiftieth of it.

The money-weighted return went from +17.64% to +17.58%, about 6 basis points lower, against 25 last week. Sleeve weights at the 16 October mark: equities 55.36% against a band of 30 to 60%, fixed income 31.59% against 25 to 50%, cash 9.45% against 5 to 20%, gold 3.60% against 0 to 8%. Nothing is breached, and fixed income remains the thinnest thing in the book relative to its band.

The largest model-book equity allocation is NMB at 18.20% of the book, down from 18.38%, 18.61% and 19.26% in the three weeks before. The single-name cap published on 1 October is 20%, and the distance to it has widened in each of the three weeks since; every time because the price fell, and not once because a decision was taken. The top two allocations together are 35.66%, essentially unchanged from 35.69%.

The status is the same as it was, and saying so plainly is the point of writing it down again: the cap binds on new savings only, it never forces a sale, and no new savings arrive before 1 November. Three quiet weeks is three weeks untested. A rule that has not had an opportunity to bind has not been vindicated by anything.

There is a sharper version of the concentration question that this week raises and the cap does not answer. The book's two bank allocations are 26.06% of the book, and they are also the two counters that supplied 91.2% of the market's turnover. That is not two risks; it is one risk counted twice. The names the book is most exposed to are the names whose price is set by the only real flow in this market, and a single-name cap at 20% does nothing about a pair that is 26% and correlated by construction. Chapter 21 considered a pair cap on 1 October and declined it, with its reasons given. This week is the strongest evidence yet for revisiting that, and the right place to revisit it is the 1 November allocation, where it can be written as a dated rule rather than improvised into a week with no money in it.

No Treasury bill auction settled in this window. The most recent remains auction 1208 of 7 October: 35-day 3.36%, 91-day 3.4908%, 182-day 5.63%, 364-day 6.2839%. The question Chapter 23 put; whether the 35-day rung's 84 basis point jump holds or gives itself straight back; is therefore still open, and will be answered by the next auction on the fortnightly cadence, before the November allocation. Against September's 4.3% inflation the arithmetic is unchanged: the 35-day pays about 0.94 percentage points less than inflation and the 91-day about 0.81 less; the 182-day about 1.33 more and the 364-day about 1.98 more. The policy rate remains 6.25%, for Q4 2026.

What did appear is a 25-year figure, and how this book handles it matters more than the figure. The Bank of Tanzania curve store's latest-reading-by-tenor list now carries 25 years at 10.84% dated 15 October, inside this window. It would be the long end's most interesting print in months: the last 25-year auction this book's per-auction evidence series holds is 11.89% on 24 June 2026, so taken at face value that is roughly 105 basis points lower at the very long end.

The Ghost is not taking it at face value. That reading carries no auction number, no source locator and no row hash, and it does not appear in the per-auction series at all; the series where every entry back to 2023 carries all three. Auction 1208, by contrast, carries its number, its endpoint, its locator and its hash. These two figures are not the same kind of fact, and this book's rule is that the Brief and the Ghost carry the same figure at the same confidence. So it is published here as unconfirmed, pending the per-auction record: reported because suppressing it would be its own distortion, and relied upon for nothing. If it is corroborated before 1 November it becomes real evidence for the duration question. If it is not, it was a row in a summary table and the Ghost will say so.

The official fitted sovereign curve, run date 13 October, twisted rather than shifted. Against the 9 October run: the six-month point rose about 30 basis points to 4.79% and the one-year about 7 to 6.25%, while the two-year fell about 20 to 8.32% and the twenty-five-year fell about 20 to 11.17%. The front end repriced up, the long end down. That is the same shape auction 1208 drew a week earlier, and the fitted curve is a model of the term structure rather than a transaction, so it corroborates a direction and settles nothing.

None of which touches the marks. The Ghost's fixed-income sleeves accrue at the indicative yields they were entered at: T-bonds at 11%, bills at 10%, the collective income fund at 6%, cash at 5%. A curve moves the hurdle for the next shilling, not the carrying value of shillings already committed. That is precisely why the duration question keeps being deferred to an allocation date instead of being traded in a week that has no money to allocate.

October's TSh 7.0m of savings is still not recorded in the canonical tracker. The public book as at 16 October carries contributed capital of TSh 521.0m; July, August and September only. This is the third consecutive weekly note to disclose it, now sixteen days after the allocation date, and Chapter 21 carries its own dated 3 October correction saying the same thing: that chapter's allocation is a published proposal, not a recorded contribution, and TSh 528.0m is the capital the proposal would produce rather than capital the book holds.

What changes when it is entered, stated in advance for the third time so nobody has to take it on trust afterwards: contributed capital becomes TSh 528.0m; market P&L is unchanged at TSh 90.03m; and the money-weighted return moves by roughly two thousandths of a percentage point. A book whose reported return barely notices new money is a book measuring return correctly.

Three weeks of an unreconciled line in public is not a good look, and it is still the better of the two available options. The alternative is to stamp a deposit with a date it did not happen on, which would make every figure above it unverifiable. Performance is never back-dated here. When the contribution is recorded it will be recorded on the date it is recorded, the sleeve weights will change on that date and not before, and this ledger will say which date that was.

Hold. Nothing trimmed, nothing sold, no sleeve opened or closed. Three reasons, in order.

One: there is no new money, and the only liquid exit is the problem itself. The next savings arrive on 1 November, so any change this week would have to be funded by a sale. In a market where two counters carried 91.2% of the turnover, the only holdings that can be sold at size are CRDB and NMB; which is to say, the concentration can only be reduced by selling into the week it fell, at a price set by the flow that pushed it down. A concentration that is also the market's liquidity cannot be traded out of cheaply. It can be grown out of, with new savings directed elsewhere, which is the standing plan.

Two: the week's new evidence is not evidence yet. The one genuinely new fixed-income print is a 25-year reading with no auction number, no locator and no hash. This book does not move an allocation on a figure it cannot source to the standard it holds everything else to. No bill auction landed, so the short-end question is unanswered rather than answered differently.

Three: the real question is a rule, and rules get dated. This week makes the case for a pair or bank-sleeve constraint better than any week so far: two holdings at 26.06% of the book that are also 91.2% of the market's flow, falling together for a third week while the non-banks rose. That is an allocation-policy question. It belongs in the 1 November note, written as a dated rule with a stated mechanism, and not invented on a Saturday to look decisive.

Two things the Ghost explicitly declined, published as non-decisions so the reasoning is on the record. It declined to trim either bank into a week of net foreign selling and holiday-thinned turnover. And it declined to treat the unconfirmed 10.84% as a reason to extend duration; if the long end wants this book's November savings, it can start by appearing in the auction record with a number attached.

Four dated things. The next Bank of Tanzania bill auction, on the fortnightly cadence after 1208, and specifically whether the 35-day holds 3.36% and whether the 364-day finally rises after six falls. Corroboration, or not, of the 15 October 25-year reading in the per-auction record; that single row decides whether the duration question has new evidence or none. A regeneration of the listed-company filing register, last generated 23 September and now 24 days stale, which would say whether anything was disclosed behind JATU's 14.81% fall or MUCOBA's round trip. And the 1 November allocation, which is the first occasion the single-name cap can bind on anything, and the right place for the pair question.

Two smaller marks. The Bank of Tanzania's official USD/TZS indicative mean moved from 2,631.3587 on 9 October to 2,633.1199 on 16 October, about +0.07%; a marginally weaker shilling that ends two consecutive firmer weeks, and small enough to be noise. Across 2026 the shilling is still about 7.5% stronger against the dollar. And the Mining Commission's world-gold reference was about flat for a second week at $4,196 against $4,192; the domestic buying-centre price was $3,693, about 12.0% below world, that gap being royalty, clearing and margin between the global market and the Tanzanian mine gate. The model book's gold sleeve is 3.60%, inside its 0 to 8% band, carried at contributed value and not marked, so none of that movement is in the NAV either way.

The discipline for the week is three lines. When a minority of the board falls and the index falls more, go and find out who carried the money, because that is the whole explanation. When a coverage ratio improves on an unchanged accrual, the loss got smaller and nothing got safer. And when a figure arrives without the provenance every comparable figure carries, publish it as unconfirmed and act on none of it; an interesting number is not evidence until it can be sourced.

N/AThe Ghost. Hypothetical and educational only; a model book, not real money, and never advice.