Skip to content
All library documents

Limits of Stock Price Forecasting and Long-Horizon Momentum Models

Article Quant Q&A · Author: Caprikuarius
Read the original

Summary

The document asks whether stock forecasting models can combine recent price movements with longer-term trends. It introduces ARIMA as a common time-series model and ARCH-type models as approaches to changing variance, then questions whether methods focused on recent observations capture behavior over months or years.

The answer argues that there is no universally established model for this broad prediction task. It invokes the efficient market hypothesis as a reason to doubt reliable price predictability, while noting that researchers can test many kinds of inputs and that momentum strategies use longer-horizon return information. The response is cautionary rather than empirical: it presents no backtest, comparative evidence, or specific model specification, and its broad claim about predictability does not rule out conditional signals or market-specific effects.

Key ideas

  • ARIMA models are presented as a common approach to price time series, while ARCH variants address variance dynamics.
  • The question distinguishes short-term price history from longer-term trend information.
  • The answer offers no single established model for the broad forecasting problem.
  • Momentum strategies can use returns measured over longer horizons.
  • The document gives no empirical comparison showing that any proposed input or model predicts returns reliably.

Tags

EquitiesMachine learningMomentumStatistics
Full text
# What are some good models for stock price predictions?


# What are some good models for stock price predictions?












For the fitting and forecasting of time-series data on stock price, the most frequent model I have heard of is ARIMA. ARIMA is actually conducting a regression of stock prices and residuals of stock prices in the few days prior to the current date. ARCH model and its derivatives model variances on the data in the previous few days.

However, for the stock price, in particular, I believe most traders will not only rely on the stock prices in the past few days but also on the general trends in the past few months or even years. Therefore, the behaviors of traders should not be modeled fully based on stock prices in the past few days.

Is there a well-established model that considers stock prices not only in the past few days but also on the overall trend? Or if not, is there a reason why modeling only on the stock prices in the past few days is better than modeling on earlier data?

Thanks!

## Answer by Stelios Kounis (score 1)

https://quant.stackexchange.com/a/54909

Long story short no. Also, your question is too general in my opinion. Stock prices are not predictable according to the efficient market hypothesis. However there are many models one can try, they can include whatever you can think of from weather data to traffic data to the stock price 10 years ago of the company. There are momentum-based strategies that for example see the average gains from the whole industry over a long time horizon. The possibilities and combinations are endless, trying to find a good model to make you rich is, to say the least, a long shot.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

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