An Indicator Is Not a Model
Milton Berg tracks 30,000 indicators and runs more than 2,000 models. Those two numbers get quoted together so often that most people assume they mean the same thing. They do not, and the difference is the entire method.
No momento, os artigos são publicados em inglês.
Cartas gratuitas do escritório de Milton Berg
Relatórios e notas de mercado gratuitos e ocasionais — direto na sua caixa de entrada. Sem spam.
Milton Berg is described as tracking 30,000 indicators and running more than 2,000 models. Those two numbers get quoted together so often that most people assume they mean the same thing. They do not, and the difference is the entire method.
An indicator is one measurement. A model is a coincidence of several.
A real model, three conditions
He walked through a real example on camera — the model that fired at the COVID low of March 2020. It has three conditions:
- NASDAQ 100 five-day volume at a 200-day high
- The NASDAQ down 7% over a three-day stretch
- The S&P volatility index between 45 and 60
No single condition is sufficient for this model. Volume spikes constantly. Even a sharp three-day drop can occur without marking a durable market low. A high VIX reading tells you people are frightened, which is not a forecast.
“These three lines are an indicator… we put the three together, and we have a model.”
What he refuses to include
The discipline is in what he refuses to include:
“We have many combinations of indicators that signal once precisely at a market low — but we don't include that in our models, because this is only one instance.”
A combination that worked exactly once is not evidence. It is a coincidence with a good story. And some measurements never earn a place at all: “Some of the indicators have never, ever satisfied the requirements for a model.”
The answer to the backtest objection
This is also the honest answer to the accusation that any long backtest attracts — that you can always find something that fit the past. You can. What is much harder is to find something that fit the past repeatedly, rarely, and at the same kind of moment every time — and then decline to use the ones that only fit once. Repetition does not by itself rule out overfitting, but it is far stronger evidence than a one-off, and it is the first question to ask of any claim that market timing works.
Perguntas frequentes
What is the difference between an indicator and a model?
In Milton Berg's usage, an indicator is a single measurement — a volume reading, a rate of change, a volatility level. A model is a specific combination of several indicators that, historically, has coincided with the same kind of market moment repeatedly and rarely.
What were the three conditions of the model that fired at the March 2020 low?
NASDAQ 100 five-day volume at a 200-day high, the NASDAQ down 7% over a three-day stretch, and the S&P volatility index (VIX) between 45 and 60 — all at once.
Why doesn't Berg use a combination that worked once at a market low?
Because one instance is not evidence. He says his firm has many combinations that signaled precisely at a single low and deliberately leaves them out of the models; a pattern earns a place only by repeating at the same kind of moment across history.
This article draws on MB Edge's own recorded interview with Milton Berg on August 26, 2026. Quotations are Berg's own words.
O MB Edge publica um modelo de mercado de longo prazo. Os resultados do modelo anteriores a maio de 2025 são backtests hipotéticos, não representam operações reais em nenhuma conta de cliente e não garantem resultados futuros. Este artigo é apenas um comentário educacional — não constitui aconselhamento de investimento individualizado nem recomendação de compra ou venda de qualquer ativo.
Leia as divulgações completas