Autoregressive Scatterplot Forecasting with Adaptive Trend Lookbacks

Summary

This indicator projects a short sequence of future price values using an autoregressive model with a one-bar lag. A library function estimates forecast results and variance bounds from a chosen training window. Users can display the central forecast and its upper and lower variance estimates as dots or horizontal segments, and optionally draw a line connecting the projected path. The script also reports correlation, squared correlation, forecast extremes, and the selected lookback length.

The lookback can be entered manually or selected from a set of candidate windows by choosing the one with the strongest correlation between price and time. The author recommends aligning the training period with the current trend and checking the model statistics. Examples are described for QQQ, but the document provides no quantified forecast accuracy or systematic out-of-sample evaluation. The author explicitly distinguishes the approach from machine learning and notes that training-window size can affect representativeness, making its projections and fit statistics insufficient evidence of predictive reliability on their own.

Key ideas

• The indicator uses a one-bar-lag autoregressive assessment to project future values over a selected horizon.
• Forecasts can show a central path alongside upper and lower variance estimates.
• An automatic setting chooses among candidate training windows based on price-time correlation.
• Correlation and squared correlation are displayed as model diagnostics, but do not establish out-of-sample accuracy.
• The author describes the method as statistical rather than machine learning and notes sensitivity to training-window choice.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.