Why Simple Market Signals Beat Complex Indicators

Two or three inputs, not twenty

Asked what surprised him most in decades of backtesting, Berg's answer was blunt: how often simplicity wins. The buy signals he prizes most are built from plain, easily tracked data — and their historical results have been anything but plain.

The numbers bear it out. Of the 1,749 tier-one signals in his library, 1,213 — nearly seventy percent — are built from just two or three components. Only 86 need five or more. Another 86 rest on a single underlying indicator.

Three signals from the public record

Berg walked through several examples in a 2026 interview. All figures below are historical observations, not projections:

• One component. The S&P 500 registers a 10-to-1 upside volume day three days in a row. Two occurrences since 1957 — August 3, 1984 and September 3, 2010 — followed by one-year S&P 500 gains of +20.51% and +23.36%, with interim drawdowns of −0.42% and −1.15%.
• Two components. At least 95% of S&P 500 stocks close above their 10-day average while the NASDAQ Composite's 10-day rate of change is 10% or greater. Six occurrences — August 26, 1982; May 5, 1997; March 17, 18, and 23, 2009; October 14, 2011 — followed by one-year gains ranging from +19.70% to +49.88%. In every case, the low in place before the signal held.
• Three components. The S&P 500 declines 12% or more and holds its low for two days; on the second day off the low the NYSE gains on higher volume and the S&P 500 rises at least 3%. Four occurrences — June 28, 1962; March 13, 2003; July 7, 2010; October 4, 2022 — followed by one-year gains of +30.12%, +39.17%, +28.61%, and +21.05%. The prior low held in three of the four cases and was violated by only −0.24% in the fourth.

A deviation-from-trend model in one sentence

Berg's favorite illustration of simplicity is a deviation-from-trend model on the S&P 500 Equal-Weighted Index — which, he notes, has historically yielded more dependable momentum signals than the familiar capitalization-weighted version. The rule: when the ratio of the index's 6-day rate of change to its 18-day rate of change reaches 105% or higher, that is a buy signal.

That single ratio has signaled 17 times, and in all 17 cases the corrective or bear-market low already in place was never broken. The median maximum gain within the following 12 months was +32.51%. One ratio, one threshold — no neural network required.

Why not MACD, RSI, and the standard toolkit?

The common oscillators — MACD, RSI, Bollinger Bands, ADX — share two properties Berg deliberately avoids. First, they are derived mechanically from price, while his inputs are mostly non-price data: volume, breadth, sentiment surveys, new highs and lows, day counts, and measures of realized and implied volatility. Second, they are always on, generating overbought and oversold readings through every market environment, which guarantees a constant stream of signals with no special claim to rarity.

Berg's models are not expected to be active — or right — in every market cycle. Each is tuned to specific conditions, factors, or regimes, and most stay silent for years. What he relies on is the combined behavior of hundreds of such models, not any single indicator that must always have an opinion.

Simplicity is a robustness strategy

A signal with two components and seven historical occurrences can be inspected, understood, and trusted — or rejected — on its merits. A model with hundreds of parameters fitted to the same seven events is, in most cases, an exercise in well-dressed overfitting.

The honest caveat is sample size: no two- or three-instance signal proves anything on its own. Berg's answer to that is architectural rather than statistical — demand rarity from each signal, then require a cluster of independent signals before treating any turn as real. Simplicity keeps each building block auditable; the cluster provides the evidence.