Why We Removed Brazil (SAO) From the DCF Discount Comparison
We previously published a DCF discount backtest for Brazilian stocks reporting 18.05% CAGR. The headline was substantially inflated by broken split-adjusted prices in our upstream FMP data. We retract the original number and explain why.
We previously published a DCF discount backtest for Brazilian stocks that reported 18.05% CAGR vs 8.58% for the Bovespa, framed as the highest CAGR of any market we tested. We've retracted that result. The headline number was substantially inflated by broken price data in our upstream provider, and we can't tell you what an honest figure would be.
Contents
- What went wrong
- Why we can't fix it locally
- What this changes for our other strategies
- Currency was always going to compress this
- Does the strategy work in Brazil?
- Read the rest of the comparison
This page replaces the original post.
Retraction published May 2026.
What went wrong
Our backtests run on FMP's stock_eod table, which serves split-adjusted historical prices via an adjClose field. For Sao Paulo (SAO) listings, that adjustment is broken on a meaningful number of stocks.
A direct query against the data showed 20+ Brazilian stocks where the maximum and minimum adjClose values differ by more than 1,000x in 2007 alone. The most extreme cases:
- CTNM3.SA:
adjCloseranges from 37.90 to 132,118,525. A 3.5 million-times ratio. - CEDO3.SA: 880,395x ratio.
- CGAS3.SA: 701,818x ratio.
- LUXM3.SA: 147,122x ratio.
These aren't real returns. They reflect reverse splits and consolidations that FMP failed to apply retroactively to the historical price series. When a backtest computes a return between two such prices, the result is fictional.
In our run, this produced individual-period returns that violate basic plausibility. One affected position contributed a 3,250% single-year return in 2013. That single artifact was enough to lift the headline CAGR by roughly 5-6 percentage points. Once you remove it (and the others like it), what's left is a number we can't validate.
The same root cause produced fake returns for Australia. That retraction is here.
Why we can't fix it locally
The error is in the source data, not in our backtest code. The adjClose field in FMP's response for these symbols is identical to the raw close even on dates where reverse splits or consolidations clearly occurred. We can filter individual symbols out, but the universe of affected names changes by year, and we can't be confident we've caught every case without reference data we don't currently have.
We've filed a bug report with FMP documenting the affected symbols.
What this changes for our other strategies
Every backtest in our strategies series that previously reported a Brazilian (SAO) result has been removed or pending retraction. For the DCF Discount comparison specifically, the global comparison post now covers 14 exchanges and explicitly excludes Brazil.
We are leaving open the possibility that the real Brazilian return on this signal is still strong. The Bovespa over 2000-2025 was a high-return index in BRL terms. Brazilian equities historically trade at depressed valuations, and FCF yield as a signal has worked across most markets where the data is clean. We just can't measure it with what we have today.
Currency was always going to compress this
Even if the headline 18% CAGR had been real, it was in BRL. The Brazilian Real depreciated from roughly 1.95 per USD in 2000 to 5.50 in 2024. That's a ~5% annual drag for a USD-based investor. The ~13% USD-equivalent figure we previously reported is also retracted, because it was derived from the inflated BRL number.
Does the strategy work in Brazil?
We genuinely don't know. The signal (FCF yield above 8.78%) is exchange-agnostic. There's no a-priori reason to expect Brazilian value stocks to behave fundamentally differently from Mexican or Indian value stocks, but we won't publish a number we can't trust.
If you want to test this yourself with a different data source, the screen is open and the methodology is documented in the US flagship post.
Read the rest of the comparison
The cleaned 14-exchange comparison and the regional posts are here:
- DCF Discount: Global Exchange Comparison (14 exchanges)
- DCF Discount: US Stocks (flagship post with full methodology)
- DCF Discount: Canadian Stocks (TSX)
- DCF Discount: Swedish Stocks (STO)
- DCF Discount: Indian Stocks (NSE)
Original post (now retracted) reported 18.05% CAGR over 2000-2025 vs 8.58% for the Bovespa, with a +9.47% annual excess. We retract those figures. Data quality details: FMP stock_eod has broken reverse-split adjustments on multiple SAO symbols, producing physically impossible 1,000x+ price ratios. Bug report filed March 2026. Not investment advice.
