What Is Turning Point Analysis? Milton Berg's Method, Explained

The core premise

Berg's starting observation, repeated across five decades of research, is that day-to-day price movement is essentially random. No indicator reliably forecasts an ordinary Tuesday. But at important tops and bottoms something different happens: many independent measures of market behavior hit extreme readings simultaneously.

The raw material is not chart patterns but market internals: abnormal spikes or collapses in volume and volatility, rates of change, TRIN, counts of new highs and new lows, the balance of down days to up days, upside and downside gaps, and advance-decline data. When several of these reach historic extremes together, the probability of a durable inflection rises sharply. Berg calls those moments turning points, and the discipline of finding them Turning Point Analysis.

Rarity is the criterion

How do you decide which readings matter? Berg's answer: build a histogram for every indicator, then ignore the crowded middle of the distribution.

Take one example he has made public — 5-day net NYSE upside volume as a percentage of total NYSE volume. Across 22,169 observations, readings range from −75% to +75%, but 89% of the time they fall between −30% and +35%. That zone is noise, and Berg largely ignores it for signal design. The work concentrates on the tails: the 5.74% of readings below −30% and the 5.26% above +35%.

The logic is simple. An indicator that speaks constantly tells you nothing. One that has spoken a handful of times in seven decades — and almost always near the same kind of market condition — is worth listening to.

Combinations, not oscillators

A single extreme is interesting; a rare combination of extremes is actionable. Of the 1,749 tier-one signals in Berg's library, 1,213 are built from just two or three components. Only 86 use five or more.

One public example: the S&P 500 printing a 10-to-1 upside volume day three days in a row. That has happened exactly twice since 1957 — on August 3, 1984 and September 3, 2010. In those two instances the S&P 500 gained +20.51% and +23.36% within a year, with interim drawdowns of just −0.42% and −1.15%. Two instances demonstrate rarity, not certainty — which is precisely why no single signal is asked to carry the weight alone.

This is also why you will not find MACD, RSI, or Bollinger Bands in the toolkit. Those indicators are derived mechanically from price and are built to always have an opinion, generating overbought and oversold readings straight through bull and bear markets alike. Berg's models are the opposite: each is tuned to specific, rare conditions, expected to stay silent for years, and judged by the combined behavior of a diversified set of hundreds of models rather than by any one indicator's constant chatter.

Tested, not curve-fit

The dataset begins on March 4, 1957 — the birth of the modern S&P 500 — with market data sourced from Ned Davis Research and Bloomberg, and custom historical series built by NDR's research team.

Berg is careful to distinguish his process from conventional backtesting. Nothing is being parameter-fitted to price patterns, moving averages, or crossovers. The process runs in the opposite direction: identify extreme readings in the underlying data first, then ask which extremes — alone or in combination — cluster near significant market turning points rather than scattering randomly across ordinary conditions.

Clusters, not lone signals

At real turning points, signals arrive in swarms. Around the April 2025 low, Berg's models fired two signals on April 4, two on April 7, four on April 8 — and fifty-seven on April 9 — with more arriving through the end of the month. Collectively, that cluster corresponded to more than 45 prior market dates, giving the composite reading a far broader historical foundation than any individual signal's three or four precedents.

That is the answer to the obvious statistical objection. Any one signal has a tiny sample size. A cluster of independent signals, each rare on its own, converging in the same week, is a different order of evidence.

From analysis to action

For institutional clients, the output is multiple primary and confirming signals per week, across asset classes, sometimes intraday. For individual investors, the same research is compressed into one binary decision: the MB Edge model's switch between 100% S&P 500 exposure and Treasury bills — a long term hypothetical model built on the same turning-point library.

Different vehicles, same premise: act when the data is genuinely extreme, and stay humble the rest of the time.

Frequently asked questions