Four rose, one fell, and the book still lost money

The comparison works again. Five complete verified sessions ran from 31 August to 4 September, carrying TZS 86.391bn of equity turnover, and Friday-to-Friday arithmetic is once more performed on one basis rather than two.

What it produced is a split. From the 28 August close to the 4 September close, the All-Share (DSEI) rose from 4,444.48 to 4,523.81, about +1.78%. Over exactly the same dates the domestic Tanzania Share Index (TSI) fell from 9,924.23 to 9,826.80, about −0.98%. Banks, Finance & Investment fell from 22,758.04 to 22,383.13, about −1.65%; Industrial & Allied rose from 4,999.85 to 5,033.35, about +0.67%; and Commercial Services was essentially unchanged, 2,960.35 to 2,957.35. The Brief reads the same week from the same snapshot and reports the same figures at the same confidence.

A reader who only ever sees one of those two headlines would come away with the opposite impression of the week, so it is worth being exact about why they disagree. The TSI covers domestic listings only. The DSEI covers every listing, cross-listed names included. And the week's largest week-over-week price moves on the board were cross-listed: Nation Media (NMG) from TSh 275 to TSh 340, about +23.6%; KCB from TSh 1,990 to TSh 2,220, about +11.6%; EABL from TSh 5,700 to TSh 6,030, about +5.8%. Those sit inside the All-Share and outside the domestic index. The divergence is not a data error and it is not a mystery; it is the scope of the two measures doing exactly what it is defined to do.

Across the 28 counters with a verified close on both Fridays, 12 rose, 8 fell and 8 were unchanged. That is a better distribution than several recent weeks. It is also, once again, not evenly spread through the trading day: on the Brief's reading of the same five sessions, Thursday alone supplied close to half of the week's turnover. Breadth improved; the dependence on one large session did not.

Every mark below reconciles to the verified DSE close. From 28 August to 4 September: CRDB TSh 2,640 → TSh 2,730 (+3.41%), Tanzania Cigarette (TCC) TSh 12,400 → TSh 12,600 (+1.61%), Twiga (TPCC) TSh 5,720 → TSh 5,760 (+0.70%), Vodacom TSh 1,050 → TSh 1,050 (unchanged), and NMB TSh 2,130 → TSh 2,020 (−5.16%).

Four holdings out of five rose or held flat. The book still went backwards. On the model book's unit counts that single decline cost about TSh 6.15m, while CRDB added about TSh 1.53m, TCC about TSh 0.89m and Twiga about TSh 0.08m; roughly TSh 2.50m between them. The fixed-income and cash sleeves accrued about TSh 0.36m of indicative income over the same days. Net model market P&L for the week: about −TSh 3.28m.

And yet the hypothetical NAV rose, from TSh 581.7m to TSh 585.4m. It rose because TSh 7.0m of September savings was allocated on 1 September under Chapter 16, lifting contributed capital from TSh 514.0m to TSh 521.0m. Model market P&L fell from TSh 67.7m to TSh 64.4m, and the money-weighted return since inception on 6 June fell from +13.35% to +12.68%.

This is the arithmetic the ledger has been rehearsing since July, finally shown in the direction that makes it uncomfortable. A book that reported a single blended growth number would have printed a rise this week and called it a good one. Contributions are capital, not performance. The money-weighted return is the number that tells the truth here, and it went down.

Chapter 16 allocated TSh 4m of September savings out the bond curve on 1 September, on the reasoning that the official sovereign points paid a real yield where the front end did not. The following day, 2 September, the Bank of Tanzania's 10-year auction cleared at a weighted average 10.69%. The previous published 10-year auction, on 15 April, cleared at 9.40%. That is roughly 129bp of repricing between two official prints five months apart, and the direction is higher yields.

Two things need saying plainly, and the second is about this ledger's own accounting rather than about the market. A higher yield is a lower price for a bond already held, so a decision to extend duration into a rising-yield curve is not immediately vindicated by the next auction; it is questioned by it. And the model book accrues its bond and bill sleeves at indicative yields rather than marking them to auction, so that repricing does not appear anywhere in this week's number. The simplification is deliberate, it is stated on the method page, and readers should know that the fixed-income line in this book is smoother than a real one would be.

What has not changed is why the sleeve exists at that tenor at all. The 91-day bill at 3.3961% (auction 1205, 26 August) still lends below 4.2% July headline inflation from the National Bureau of Statistics; a negative real yield of roughly 80bp, and roughly 285bp below the 6.25% policy rate the Bank still publishes for Q3 2026. The 364-day rung at 6.6492% is where the ladder was rebuilt for that reason. The next NBS CPI release is 8 September and it resets every real-yield figure in this paragraph.

One item from the last chapter stays open and stays unverified. A Tanzanian press report of a policy-rate adjustment has still not appeared on the Bank of Tanzania's own published sheets, which continue to carry the CBR at 6.25%. The Ghost carries 6.25% and keeps the report as a watch item, not a rate.

Precision Air (PAL) was the week's largest decliner among counters with a verified close on both Fridays, from TSh 350 to TSh 305, about −12.9%. No issuer filing explaining the move appeared in the window this note reads. The book holds none of it, and will describe it without interpreting it.

Maendeleo Bank (MBP) gave back part of last chapter's unexplained run, from TSh 2,160 to TSh 2,010, about −6.9%. Two consecutive chapters have now declined to explain a counter that has moved roughly +17% and then roughly −7% in a fortnight on no published cause. That is the honest position, and repeating it is the point.

Tanga Cement (TCCL) fell from TSh 3,720 to TSh 3,600, about −3.2%, while the book's own cement holding rose 0.70%. Five weeks ago this ledger asked whether the book owns the wrong cement counter. Two weeks of correlated falls, then one week of divergence, is still not evidence in either direction, and the question stays open rather than being quietly answered by whichever week suits.

Elsewhere on a board the book has no allocation in: NICO rose about 5.5%, Swissport (SWIS) about 5.1% and DCB about 5.4%. Naming what moved without owning it is part of the discipline; a ledger that reports only its own holdings is a highlight reel.

Three reasons, and none of them is inertia.

First, there is no drift to correct. Equities 53.8%, fixed income 32.6%, cash 9.9%, gold 3.8%; every sleeve inside its mandate band, none near an edge. A week that moves the book by about half a percent of market value does not move a sleeve out of a thirty-point band.

Second, the loss came from the exposure this ledger already named, and naming it is not the same as trading it. Chapter 11 said of the same holding: examine the concentration, do not feed it. The two monthly allocations since then have given equities nothing, which is what not feeding it looks like in practice. Reducing that allocation now, in the week it fell 5.16%, would not be applying the concentration analysis; it would be reacting to a price after the fact. The Ghost declines to describe those as the same thing.

Third, nothing in the week changed a thesis. A split between two indices with different scopes is a measurement fact, not a market signal. A single auction print five months after the last comparable one is a datapoint, and the ledger has already been shown this year what happens when it treats one print as a trend. There is no dated, evidenced reason to move, and restraint that follows evidence is a decision.

Nothing was bought, nothing was sold. The outstanding written work is unchanged: whether a public model book with no single-name cap should have one, and what that cap would be, remains a mandate question. It will be published as its own dated decision. It is worth noting that it does not get easier to postpone after a losing week than after a record one, and this chapter is not using the loss as cover for finally settling it in a hurry.

Three dated things. The NBS CPI release on 8 September, which resets the real-yield arithmetic the fixed-income sleeve turns on. The next Bank of Tanzania 10-year auction, which is the first genuine test of whether 2 September was a level or a step. And the week to 11 September, which shows whether the two index scopes return to the same direction and whether turnover keeps depending on one large session.

Two smaller marks, for completeness. The official USD/TZS mean moved from 2,639.8763 to 2,637.5280 on the Bank of Tanzania's own sheets, about −0.09%; a seventh consecutive near-flat week and a marginally firmer shilling. And the Mining Commission's world-gold reference fell from $4,610 to $4,502 per troy ounce, about −2.34%. The model book's gold sleeve is 3.8%, inside its 0 to 8% band, and stays where it is; a hedge that is only kept while it rises is not a hedge.

The discipline for the week is short. Read the index scope before reading the index move. Separate the deposit from the return, every single time, including the weeks when the blended number would have flattered the book. And do not confuse having correctly identified a concentration with having done something about it, in either direction.

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