The research · filed 26 Oct 2026, 14:22, before we bought
Why we bought these ten companies
Written by Tavrise AI on the day of the post-election crash, reviewed by a person, and sealed before the first order. About a 12-minute read.
The thesis
The market had positioned for the other candidate, so the result did two things at once. It priced out that bet, which is rational. It also forced indiscriminate selling of liquid Brazilian stocks, which is not: high-return companies with net cash fell as much as weak ones, priced as if a fiscal break were already certain.
The lowest-risk way to capture a partial normalisation is to own those companies, including the sector that benefits directly from the winner’s policy (low-income homebuilders), with a dollar-earning position as cover if the pessimistic fiscal case happens.
What happened on 26 October
| Item | What happened |
|---|---|
| Result | Lula 51.79% against Flávio Bolsonaro 48.21%; the first time a second-place finisher after round one won |
| Ibovespa | From 205,083 on Friday to about 173,000 (−15.5%); low of 169,516 (−17.3%); circuit breaker at 9:57, the first since March 2020 |
| Positioning | The index had been up 27% for the year and 10% in October: the market was priced for the other result |
| Small caps | −27% |
| Dollar | From R$ 4.99 to a peak of R$ 5.60 (+12%) |
| Interest rates | Long DI futures up about 190 basis points, to around 14.5% |
| Treasury | Cancelled the week’s auctions and bought back about R$ 7.1 billion of fixed-rate bonds |
The pattern was a textbook risk-off unwind: the dollar up, long rates up, and anything domestic, liquid and owned by foreigners or index funds sold together. Exporters rose (Suzano +10%, Embraer +8%, WEG +5%, Vale +5%) and defensives fell least, while some of the best domestic companies fell the most: B3 −36%, Lojas Renner −34%, BTG Pactual −31%, Multiplan −28%, Direcional −24%, Cury −21%, Sabesp −21%, Rede D’Or −18%. The index fund BOVA11 fell 14.2%.
The benchmark problem
We only call this a win if it beats both the Ibovespa and cash (CDI). In Brazil, cash is a very high bar: about 14% a year from here. Since 2000, the Ibovespa itself beat cash in only about half of all 12-month periods.
| Starting point | 12-month periods | Ibovespa beat CDI | Median excess |
|---|---|---|---|
| Any day since 2000 | 6,393 | 49% | −1.1 pts |
| The day after a shock (−5% in a day, or −15% from the 60-day high) | 22 | 50% | +0.4 pts |
| Single-day falls of 8% or more (2008, 2011, 2017, 2020) | 4 | 4 of 4 | +12 to +33 pts |
The election analogues are the uncomfortable part:
| Runoff | Next day | Ibovespa, 12 months later | CDI, 12 months |
|---|---|---|---|
| 2002 (the sell-off came before the vote) | −4.4% | +87.8% | +24.2% |
| 2006 | −1.1% | +63.4% | +12.5% |
| 2014 (the market bet on the opposition and lost) | −2.8% | −5.5% | +13.1% |
| 2018 | −2.2% | +29.4% | +6.3% |
| 2022 | +1.3% | +1.8% | +13.5% |
The two cases most like this one, 2014 and 2022, lost to cash by more than 10 points. But this shock was far larger than any of those (−15% in a day), so far more was already priced in. That is both the opportunity and the warning.
Why a defensive portfolio was the wrong tool
Write a portfolio’s return as its stock-picking edge plus its sensitivity to the market (beta) times the market’s move. A defensive basket with a beta of 0.8 must out-pick the market by 3 to 5 points a year in every scenario to beat both benchmarks. A basket with a beta near 1.2 beats the Ibovespa whenever the index rises, and beats cash whenever the index rises more than about 12%, before any stock-picking edge. For this goal, low risk mostly means losing less to cash.
How we chose the companies
We screened all 370 listed stocks with live prices from 14:22 for companies that:
- fell 18% or more that day;
- trade at least R$ 30 million a day;
- trade at 15 times earnings or less;
- earn at least 15% on their equity;
- have net debt below one year of operating cash earnings (EBITDA), or are financial companies.
Six passed: B3, BTG Pactual, Cury, Direcional, Tenda and Vivara. We then applied judgement: allowing inflation-linked regulated businesses with moderate debt (Sabesp, Multiplan), net-cash Lojas Renner despite a 14% return on equity, and a premium for Rede D’Or’s quality; capping homebuilders at 17.5%; keeping Itaú as the core bank and Vale as the dollar hedge.
We excluded anything state-controlled or highly indebted, however cheap: Petrobras, Banco do Brasil, Equatorial, Localiza, Cosan and Cyrela among them. A government in its fourth term has specific losers (state companies, privatisation stories, oil exporters, heavily indebted long-duration businesses) and specific winners that were sold anyway: low-income housing, funded mainly by FGTS at subsidised rates, and mass-market consumption helped by the income-tax exemption up to R$ 5,000.
The portfolio we bought
| Company | Weight | That day | P/E | ROE | Role |
|---|---|---|---|---|---|
| ITUB4 | 15% | −14% | 9.8 | 22% | The best-run private bank at under 10 times earnings, with a dividend near 8% |
| VALE3 | 12.5% | +5% | n/a | n/a | Dollar revenue as a hedge; the largest Ibovespa weight keeps us close to the index |
| B3SA3 | 10% | −36% | 14.3 | 28% | The exchange monopoly with net cash; volumes rise in crises |
| BPAC11 | 10% | −31% | 11.6 | 22% | The highest-return large bank, priced for the worst |
| CURY3 | 10% | −21% | 7.2 | 66% | Builds for the housing program the winner prioritises; net cash |
| LREN3 | 10% | −34% | 6.5 | 14% | Net-cash apparel leader at about 7 times earnings; mass-income policies help |
| SBSP3 | 10% | −21% | 11.0 | 18% | Privatised water utility with inflation-linked tariffs |
| DIRR3 | 7.5% | −24% | 5.3 | 37% | A second housing-program builder, to spread execution risk |
| MULT3 | 7.5% | −28% | 10.0 | 21% | Premium malls with inflation-linked rents |
| RDOR3 | 7.5% | −18% | 16.6 | 23% | The largest hospital network; long-term demand |
Financials 35% (one traditional bank), homebuilders 17.5%, mining 12.5%, retail 10%, sanitation 10%, malls 7.5%, healthcare 7.5%. Multiples are trailing provider figures at the 14:22 price, checked against company filings at each review.
Risk
| Portfolio | On 26 Oct | If prices return to Friday | Beta, 5 years |
|---|---|---|---|
| Quality & Value | −21.1% | +29.5% | 1.21 |
| Index fund (BOVA11) | −14.2% | +16.6% | 1.00 |
| The first proposal we rejected | −7.2% | +8.6% | 0.76 |
- It moves about 1.2 times as much as the market. If the Ibovespa falls another 15%, expect about −18%.
- Its weekly correlation with BOVA11 is 0.90–0.92: ten names are already diversified, and more would pull it toward the index.
- If history repeats 2014 or 2022, it loses to cash by more than 10 points.
- A 15% loss triggers a full documented review, never an automatic sale.
What would prove us wrong, and when we will know
The thesis is wrong if:
- the new economic team lacks credibility and long-term rates stay above 15.5–16% into 2027;
- a credit cycle hits banks and retailers;
- funding for the Minha Casa Minha Vida housing program is cut.
We will know mostly from the Finance Minister appointment (weeks), the first Copom decision after the election, the 2027 budget and fiscal target (one to three months), and third- and fourth-quarter results. Each company also has its own written sell rules, on its page.
The first proposal we rejected
Before this portfolio, a different AI model proposed a low-risk basket of ten companies that had barely fallen or had risen with the dollar (Vivo, Porto, Ambev, Suzano, WEG and others), with tight price stops and early profit-taking. Tavrise AI reviewed it adversarially and rejected it:
- it answered a different question (what low-volatility basket to own on any day) from the one we asked (how to capture a recovery);
- with a beta of 0.76, it needed to out-pick the market by 3 to 5 points a year just to beat both benchmarks, and its own base case was below cash;
- stops 6–8% below entry, in a market that had just moved 15% in a day, would likely have sold at the bottom;
- its conditional entries would have made the track record impossible to score.
That is how Tavrise AI is meant to work: several models, each checking the others, and a person deciding.
Data and sources
Prices from brapi (live snapshot at 14:22 on 26 Oct, cross-checked with a second source to within 0.25% for all 66 tickers used); fundamentals from brapi’s company data; the Ibovespa since 2000 and daily CDI from the Banco Central (series 12).
- TSE: election result
- InfoMoney: Ibovespa live, 26 Oct 2026
- Reuters via Investing.com: circuit breaker
- Yahoo Finance: Brazilian stocks plunge
- BPMoney: homebuilders fall
- Seu Dinheiro: JPMorgan on housing-program builders
- Terra: what to expect from the economy
- Banco Central: CDI, series 12
Research as filed on 26 Oct 2026, lightly edited for reading. Figures are as of that day. Not investment advice.