Four Oscillators, One Opinion: The Redundancy Problem
Posted: Tue Oct 27, 2026 11:58 pm
A chart carrying RSI, stochastic, CCI and MACD can look like four independent confirmations. In practice, all four are functions of the same input: recent closing prices. Each transforms that one series with different smoothing, so they tend to move together, and agreement between them is largely guaranteed by construction.
Information theory offers a useful idea here. Confirmation is valuable only when a second source brings facts the first did not contain. Two oscillators computed from the same closes over similar windows share most of their content, so the second adds very little new evidence. What it does add is confidence, which is a different thing, and the extra confidence can lead to larger positions and fewer questions about the setup.
Further, each added condition reduces the number of qualifying trades, and with fewer samples a backtest becomes easier to overfit. A more productive mix draws on different kinds of information: a trend measure, a volatility measure, and something separate from price altogether, such as session timing or spread behaviour. Testing whether removing one indicator changes results is a simple check, and if results hardly change, that indicator was likely contributing decoration rather than evidence.
Information theory offers a useful idea here. Confirmation is valuable only when a second source brings facts the first did not contain. Two oscillators computed from the same closes over similar windows share most of their content, so the second adds very little new evidence. What it does add is confidence, which is a different thing, and the extra confidence can lead to larger positions and fewer questions about the setup.
Further, each added condition reduces the number of qualifying trades, and with fewer samples a backtest becomes easier to overfit. A more productive mix draws on different kinds of information: a trend measure, a volatility measure, and something separate from price altogether, such as session timing or spread behaviour. Testing whether removing one indicator changes results is a simple check, and if results hardly change, that indicator was likely contributing decoration rather than evidence.