Linear regression in price forecasting: what it can and cannot do

Why log prices\nPrices compound, so fitting a straight line to the logarithm of price measures a constant percentage growth rate rather than a constant rupee change.\n\n## How MarketLens uses it\n- A least-squares line is fitted to recent log closes.\n- The slope is extrapolated to the target date.\n- Its historical accuracy is measured with walk-forward testing before it receives any weight.\n\n## Limits\nTrends end abruptly. Linear regression has no notion of reversals, news or volatility clusters, which is why it is only one member of an ensemble and why its weight falls automatically when it underperforms.