The index broadened. The money still didn't.
Friday to Friday; 7 August to 14 August; the All-Share rose 1.18% to 4,232.85 and the domestic Tanzania Share Index rose 1.77% to 9,331.91 (DSE, 14 Aug verified close). The sectors agreed for once: Banks, Finance & Investment rose 2.19% to 20,714.64, Industrial & Allied 1.33% to 5,088.19, and Commercial Services was effectively flat at +0.05%. Thirteen of 28 comparable counters rose; the best breadth reading the book has recorded in a month.
More money turned up, too. The five verified sessions traded TSh 9.75bn, 3.96bn, 4.83bn, 4.26bn and 7.25bnUnavailableTSh 30.05bn in total, against TSh 25.77bn the week before. And still, in every one of those five sessions, the leading counter absorbed between 49% and 76% of the value that changed hands. More money arrived at the same few doors. That is a better week than last week and the same structural fact as last week, and it is worth being precise about which is which.
The book's own marks, 7 August to 14 August, are where the week actually differed. CRDB went TSh 2,590 → TSh 2,700 (+4.25%). Twiga (TPCC) went TSh 5,850 → TSh 6,050 (+3.42%). TCC went TSh 12,300 → TSh 12,500 (+1.63%). NMB went TSh 17,580 → TSh 17,700 (+0.68%). Vodacom did not move at all, closing at TSh 1,020 for the second Friday running. Every mark reconciles to the verified DSE close.
The book ended at TSh 560.6m, +9.21% money-weighted since inception on 6 June, against TSh 514.0m of contributed capital; TSh 46.6m of model market P&L. The week added roughly TSh 4.2m, of which about TSh 3.8m came from equity marks and the remainder from the income sleeves' accrual. Split by contributor, the equity move was roughly 49% CRDB, 23% TCC, 17% NMB, 11% Twiga and nothing from Vodacom. For two weeks running a single holding had supplied 85% and then 92% of the book's week. This week the two largest allocations between them supplied about a sixth of it, and the work was done by the third, fourth and fifth. That is the thing the last two notes said they wanted to see.
A seventh week mid-band
Equities 53.2% (band 30 to 60), fixed income 32.9% (25 to 50), cash 10.1% (5 to 20), gold 3.8% (0 to 8). Nothing has drifted far enough to force a decision. A week that finally looks the way you hoped is not a reason to invent one.
The fact is unchanged; the week's evidence isn't
NMB (17.6%) and Vodacom (15.2%) are still 32.8% of the model book; the same number as last week. But neither carried this week's gain. One week of broader contribution does not retire a structural question about single-name weight; it just means the question was not answered this week either way.
The front end printed; in two directions
Bill auction 1204 (12 August) is the first new front-end print since 15 July: 35-day 2.1276%, 91-day 3.4390%, 182-day 5.1803%, 364-day 6.7441%. Against auction 1203 that is the 182-day +40bp, the 91-day +9bp and the 364-day −28bp. The belly paid more; the one-year paid less.
The duration idea gets a fifth week on the table, dated and unexecuted
Since mid-July the book has carried one written, dated idea it has not acted on: extend duration while the long end pays and the front end sits pinned below inflation. For four weeks the front end refused to print, so there was nothing to test it against. On 12 August it printed.
What it said was mixed, and the mixture matters more than the direction. The 364-day weighted average yield came in at 6.7441% against 7.0258% at auction 1203 on 15 July; 28bp lower. That is the second consecutive piece of evidence pointing the same way: a week earlier the 25-year printed 11.48% on 5 August against 11.89% at the prior 25-year auction on 24 June, 41bp lower. Both ends of the trade the book was waiting to be paid for now pay slightly less than they did a month ago. Falling auction yields are perfectly good news for a bond sleeve the book already holds; they are a weaker argument for adding one.
The 182-day is the exception, and it is the reason not to write the idea off. It rose 40bp to 5.1803%; the belly of the bill curve is the one place that got more attractive, not less. That is a genuinely new observation and the book has exactly one auction of it. One print is a datapoint. The Ghost is recording it and waiting for the second.
Meanwhile the reason the idea exists at all has not gone away. The 91-day bill at 3.4390% sits below 4.2% July headline inflation (NBS); a negative real yield of roughly 76bp; and roughly 281bp below the 6.25% policy rate. That gap is exactly why August's new savings went out the bond curve rather than into the bill ladder, and nothing in auction 1204 changes that reasoning. The next two dated tests are already in the diary: the 15-year bond auction on 19 August and the next bill auction on 26 August.
What the book does not own, and will not explain
MCB remains the counter the book keeps describing and refuses to interpret. Two weeks ago it fell 56.8%; last week it ran +64.9%; this week it reached TSh 525 on 10 August and then fell every session to TSh 405 by 14 August, including −7.95% on the Friday, which made it the week's largest decliner. No issuer filing explaining any of the three legs appeared in the window this note reads. The book holds none of it. A price with no sourced cause gets described, never explained; and a counter that has now travelled −57%, +65% and −23% inside a month is the clearest available argument for why.
Tanga Cement (TCCL) is the other one, and it is the honest footnote to a discomfort this book aired a fortnight ago. Between 7 and 14 August it rose from TSh 3,710 to TSh 3,950, while the book's own cement holding, Twiga, rose 3.42%. Both cement counters were up. The question of whether the book owns the wrong one is not settled by a week in which they moved together, and pretending otherwise in either direction would be the sort of tidy story this ledger exists to avoid.
Elsewhere on a board the book has no allocation in: DCB rose 10.71% and NICO 4.95% across the week, and Tanzania Breweries filed a stronger company result with the exchange. Naming what did well without owning it is part of the discipline. A book that only reports the counters it holds is not a record; it is a highlight reel.
Why hold; a seventh week
Three reasons, and none of them is inertia. Every sleeve is mid-band, so there is no drift to correct. The one dated thesis on the table got its long-awaited evidence and the evidence narrowed the reward rather than opening it; acting now would be acting because data arrived, not because it pointed anywhere. And the single genuinely encouraging development of the week; that the book's gain finally came from four holdings rather than one; improved without the Ghost doing anything at all.
That last one is worth sitting with, because it is the easiest place for a model book to fool itself. The concentration the last two notes worried about did not get fixed by a decision; it got quieter for a week because CRDB and Twiga happened to move. The underlying weights are identical. Confusing a better week for a solved problem is how books drift.
Nothing was bought, nothing was sold. The outstanding written work is unchanged: whether a model book with no single-name cap should have one, and what it would be. That remains a mandate question, and mandate questions are published as their own dated decision; not folded into a weekly note because a convenient week made them feel less urgent.
The subdivision is no longer two weeks out. It is next week.
NMB's approved one-for-ten share subdivision takes effect on 24 August 2026 (issuer notice, approved 24 July, announced 27 July). The published calendar runs: 19 August last cum-split trading day, 20 to 21 August trading suspension, 21 August register closure, 21 to 23 August register update, 24 August effective with trading resuming on the new basis.
For the model book, whose largest equity allocation this is, the mark will step from roughly TSh 17,700 to roughly TSh 1,770 in a single session, with ten times the units. A subdivision multiplies the share count and divides the price by the same factor: the units change, the value and every holder's proportional ownership do not. This is a mechanical basis break, not a loss, it will be labelled as one on the day, and it will also mean the board's tape is interrupted for two sessions in the middle of next week. Anyone reading a chart across 24 August without that context will read a 90% fall that did not happen.
Two other dated events sit in the same fortnight: the 15-year Treasury-bond auction on 19 August, the next official long-end price discovery after this week's bill curve, and the next bill auction on 26 August, which supplies the second datapoint the 182-day observation needs. The next NBS CPI release is 8 September.
Two smaller marks, for completeness. The official USD/TZS mean moved from 2,644.79 to 2,645.24Unavailable+0.02%, a fifth consecutive near-flat week on the Bank of Tanzania's own sheets. And the Mining Commission's world-gold reference rose 2.66% Friday-to-Friday on the same-source series; the book's gold sleeve is 3.8%, comfortably inside its 0 to 8% band, and stays where it is.
The discipline for the week is therefore short. Watch whether the improved breadth survives a week whose middle is missing two sessions. Read the 19 August auction on its own terms. Label the subdivision correctly on the day. And hold what the thesis owns until something other than one good week argues otherwise.
UnavailableThe Ghost. Hypothetical and educational only; a model book, not real money, and never advice.