Ah I believe the difference is that with a dynamic forecasting model we make a prediction, refit the model on the data + the prediction that we have made and then make a new prediction (so our previous step forecast is used for model fitting). For multi-step recursive forecasting we simply use the same model without refitting. Correct me if I'm wrong.
@@Blaze098890 Hi, actually you are right, dynamic and multi-step recursive (or rolling) are the same. Dynamic simply means multi-step forecasting. We can also have re-estimation at each step in both static and dynamic forecasting. See an example in the case of bitcoin price forecasting at www.mdpi.com/1911-8074/12/2/103
Hi, nice explanation! I have a question, though. :) Why do you say static forecasting, normally, is better for out-of-sample? Because when you forecast out of your sample, you do not have data to use as input, so, all forecasts out-of-sample wont be dynamic?
Hi, static and dynamic are two concepts. Then, in-sample and out-sample are two other concepts (see ruclips.net/video/16IKsLlZClY/видео.html). Out-sample forecast can be both static and dynamic. In addition, we have the concepts of recursive and rolling forecasts (see ruclips.net/video/Tne1e4Scyl4/видео.html). Hope, this clears it.
@@RESEARCHHUB Yes, I understood this concept of in and out, like training and test sets used in machine learning. In fact, my question is: In the case of out-of-sample (test set) forecasting, is not "cheating" to give the observed data, present in the test sample, to the model, in order to do the static forecasting? I mean, is not static forecasting, in this way, always better than the dynamic, simply because the model will have access to more data?
@@thiagoribasbella4302 it is not cheating as we do not provide data of the next period that will be foretasted. But I see what you mean. Static will always be better than dynamic as static always forecasts only one period ahead. However, their application depends on the problem at hand. One might be interested only in next day forecast (static) and another might be in next 7 days forecast (dynamic). Based on forecasting theory, the further ahead we forecast, the higher error will be encountered.
![Summer Walker - Heart Of A Woman [Official Lyric Video]](http://i.ytimg.com/vi/SFddctbZzw8/mqdefault.jpg)
Very good explanation. Thank you very much!
Thank You. 🤩
thank you!
How does dynamic forecasting differ from multi-step recursive forecasting? Are they the same thing?
Ah I believe the difference is that with a dynamic forecasting model we make a prediction, refit the model on the data + the prediction that we have made and then make a new prediction (so our previous step forecast is used for model fitting). For multi-step recursive forecasting we simply use the same model without refitting. Correct me if I'm wrong.
@@Blaze098890 Hi, actually you are right, dynamic and multi-step recursive (or rolling) are the same. Dynamic simply means multi-step forecasting. We can also have re-estimation at each step in both static and dynamic forecasting. See an example in the case of bitcoin price forecasting at www.mdpi.com/1911-8074/12/2/103
Hi, nice explanation!
I have a question, though. :)
Why do you say static forecasting, normally, is better for out-of-sample? Because when you forecast out of your sample, you do not have data to use as input, so, all forecasts out-of-sample wont be dynamic?
Hi, static and dynamic are two concepts. Then, in-sample and out-sample are two other concepts (see ruclips.net/video/16IKsLlZClY/видео.html). Out-sample forecast can be both static and dynamic. In addition, we have the concepts of recursive and rolling forecasts (see ruclips.net/video/Tne1e4Scyl4/видео.html). Hope, this clears it.
@@RESEARCHHUB Yes, I understood this concept of in and out, like training and test sets used in machine learning. In fact, my question is: In the case of out-of-sample (test set) forecasting, is not "cheating" to give the observed data, present in the test sample, to the model, in order to do the static forecasting?
I mean, is not static forecasting, in this way, always better than the dynamic, simply because the model will have access to more data?
@@thiagoribasbella4302 it is not cheating as we do not provide data of the next period that will be foretasted. But I see what you mean. Static will always be better than dynamic as static always forecasts only one period ahead. However, their application depends on the problem at hand. One might be interested only in next day forecast (static) and another might be in next 7 days forecast (dynamic). Based on forecasting theory, the further ahead we forecast, the higher error will be encountered.
@@RESEARCHHUB Now I am satisfied!! hahaha Thank you for your time and explanation!! Nice playlist about forecasting, by the way.