Trading oscillators: compare what each number measures

Which oscillator answers which question?

The formula determines the interpretation. A value of 80 has a different meaning in RSI and Stochastic; a MACD value of 2 is in price units, not 2%.

The same close can produce a different Stochastic reading

Suppose the close is 105. In a lookback range from 90 to 110, raw %K is 100 × 15 / 20 = 75. In a range from 100 to 130, it is 100 × 5 / 30 ≈ 16.67. These invented examples use unsmoothed %K. The closing price is identical, but its position within the selected range differs.

• Changing the lookback can change the high and low even if the current bar does not move.
• If lookback high equals lookback low, this quotient has a zero denominator. Check the implementation’s handling instead of assuming a valid neutral reading.
• Compare the same feed, session, source, lookback and smoothing. A raw calculation need not match a smoothed chart line.

Positive money flow does not prove buying by a particular group

For one bar with high 12, low 8 and close 11, the CMF close-location multiplier is ((11 − 8) − (12 − 11)) / (12 − 8) = 0.5. Yet the price has fallen if the previous close was 12. A positive contribution and a close-to-close decline can coexist. This is a one-bar contribution example, not a multi-bar CMF calculation.

• CMF aggregates these multipliers weighted by volume over its window. Missing or zero total volume needs explicit handling.
• MFI uses a different construction and can disagree. Agreement is not proof of participant identity.
• An exchange’s traded volume, broker tick volume and an incomplete venue feed are different inputs. Name the available series.

Define divergence before counting it

One reproducible approach is to confirm price pivots using two completed bars on each side, then sample the oscillator on those same price-pivot bars. This differs from independently choosing oscillator pivots after seeing the outcome. A two-right-bar pivot becomes observable two bars later; drawing a marker on its earlier pivot bar does not make the information available earlier.

• Specify strict or inclusive ties, pivot spacing, missing readings and the reset rule. These choices change the event count.
• Keep the pivot timestamp and confirmation timestamp separately. Entries must use information available at their actual decision time.
• This page supplies definitions and arithmetic, not a measured divergence win rate.

Turn an oscillator reading into a testable hypothesis

For a close-based RSI recovery example, specify that the prior completed reading was at or below 30 and the current completed reading is above 30. That is a crossing event, whereas RSI above 30 is a state that can persist for many bars. Neither alone defines an entry fill, exit or position size.

• Decide whether the model uses completed bars or developing intrabar values. Do not compare the two as if their signals were identical.
• Treat a trend filter as another test parameter. A flat ADX line can occur at a high level; it does not by itself establish a ranging market.
• Set the trading universe, cost model and evaluation period before looking for favorable outcomes. Adding correlated indicators can increase complexity without adding useful information.
• Use a screener only within its supported symbols, intervals and data limits. This page does not promise a real-time scan of the entire market.