Reasonable length of forecast horizon in a time series

In summary, the conversation discusses the collection of monthly data for the inflow of a river and demands from a reservoir over the past 35 years. The focus is on predicting the downstream reservoir's ability to meet future demands after the construction of a dam upstream. The speaker mentions using a simple time series model for future predictions and asks how many years can be reasonably predicted. They also suggest using Monte Carlo simulations based on historical trends and variance for more accurate projections.
  • #1
ssd
268
6
Suppose we have monthly totals of observed data for last 35 years. That data is of inflow of a river in a reservoir and monthly demands from the reservoir. We are interested to check the effect of construction of a dam in the upstream. The effect is, whether the downstream reservoir will have enough water to meet its dependent demands after the upstream construction. We also have 12 estimated monthly demands for the proposed dam which are assumed to be fixed for future years.
For future prediction of inflow and demand, I primarily will use a simple time series model using regression for linear trend, without cyclic component. My query is, how many future years can I predict reasonably. I know that there is no hard and fast rule, but still can I find a logical length of forecast horizon from some sort of thumb rule?
 
Physics news on Phys.org
  • #2
The future inflow should not be a problem. Just compare weather conditions for the past 35 months to the full weather record. That's usually many decades and provides enough information to determine the range on annual inflows you can expect - long term effects (such as global warming) not withstanding.

To predict future demand, investigate your consumer base. Is the area you are serving about to experience a serious build-up. This information should be easily accessible through government agencies, for example those who issue building permits and who are as interested as you are in planning.
 
  • #3
If you want to project into the future, I recommend using Monte Carlo simulations based on historical trends and variance. You can run many simulations with reasonable projected random data and see what the distribution of results is.
 

1. What is the reasonable length of forecast horizon in a time series?

The reasonable length of forecast horizon in a time series depends on several factors such as the data being used, the purpose of the forecast, and the level of accuracy desired. Generally, a forecast horizon of 6-12 months is considered reasonable for short-term predictions, while a horizon of 1-5 years is reasonable for medium-term predictions. Long-term forecasts may have a horizon of 5-10 years or more.

2. Why is the forecast horizon important in time series analysis?

The forecast horizon is important because it determines the length of time for which a prediction will be made. It also affects the accuracy of the forecast, as longer horizons may introduce more uncertainty and variability in the data. Additionally, the forecast horizon helps in determining the type of forecasting method that should be used.

3. How can the reasonable length of forecast horizon be determined?

The reasonable length of forecast horizon can be determined by analyzing the historical data and identifying any trends or patterns. It is also important to consider the stability and seasonality of the data. A good practice is to test different forecast horizons and compare the accuracy of the predictions to determine the most reasonable length.

4. Can the forecast horizon change over time?

Yes, the forecast horizon can change over time. This can be due to changes in the data, such as a shift in the trend or seasonality, or changes in the purpose of the forecast. It is important to regularly review and adjust the forecast horizon to ensure the accuracy of the predictions.

5. Are there any guidelines for choosing the reasonable length of forecast horizon?

While there are no strict guidelines, some general recommendations include using a shorter forecast horizon for higher frequency data and a longer horizon for lower frequency data. It is also important to consider the level of uncertainty and the purpose of the forecast. Consulting with experts in the field and conducting sensitivity analysis can also help in determining the most reasonable length of forecast horizon.

Similar threads

I Time series and why would we remove seasonality and trend?
    • Set Theory, Logic, Probability, Statistics
Replies
10
Views
907
MHB Attempting to add logic mathematics to a college paper - need your input
    • Set Theory, Logic, Probability, Statistics
Replies
2
Views
2K
Time Series Analysis: 3x Exp. Smoothing
    • Set Theory, Logic, Probability, Statistics
Replies
1
Views
1K
I Regression Prediction with Time Series Data
    • General Math
Replies
1
Views
800
Creating a Predictive Model for Time-Independent Data Analysis
    • Set Theory, Logic, Probability, Statistics
Replies
13
Views
4K
  • Sticky
Insights Signature of the Sun: Analyzing the Solar Balmer Series Lines
    • Atomic and Condensed Matter
Replies
2
Views
4K
B Any plausible explanation for dynamical length contraction?
    • Special and General Relativity
3
Replies
83
Views
3K
I Two recent tests of loop quantum gravity theory
    • Beyond the Standard Models
Replies
6
Views
367
B A Valid Position on Time Theory?
    • Cosmology
Replies
2
Views
1K
Time loops, black hole and time dilation.
    • Sci-Fi Writing and World Building
Replies
2
Views
2K

Hot Threads

Recent Insights

Back
Top