The reason for the question is to verify the essentials of the parameters estimated to prediction and theoretical considerations.
The conceptual model upon which mainstream (neoclassical) economics bases its analyses consist of a circular flow of income between producers and consumers, mediated by means of markets. In this "perpetual motion" of interactions between firms that produce and households that consume, little or no accounting is given to the flow of energy and materials from the environment and back again, and little account given to human interactions that take place outside of market processes. Analyses by natural scientists (and others) find that the conventional model is simply not credible (see Hall and Klitgaard, 2019).
Dear Olatunji
Testing hypothesis is used to how close of estimated parameter to hypothesized one,but forecasting is used to estimate the values of fenomena in simple future or predect the value of dependnt variable when the value of X is drtrtmined.
In textbooks these are usually treated as very different exercises. It's true that classical hypothesis testing and econometric forecasting both require that the investigator have a model for the data generation process. In hypothesis testing, however, the value(s) of some parameter(s) is/are treated as open to question whereas in econometric forecasting the parameter values are accepted without question. That's a textbook answer; in practice investigation usually proceeds (rightly or wrongly) as a sequence of interactions with the data and this may blur the simple distinction argued above. For example, calculated forecasts may be so unexpected as to make the investigator suspect that some parameter values should be questioned.
It might be interesting to hear what a Bayesian statistician makes of this question.
@ Respected Vince, for we Bayesian probability is regarded as representing a degree of reasonable belief; numerical probabilities associated with the degree of confidence that the researcher has in propositions about empirical phenomena.
I agree with Olatunji A Shobande's description of the Bayesian perspective. Nevertheless, I think that Bayesians still distinguish between parameters - which is the focus for hypothesis testing, and random variables - which is the focus for prediction.
Econometrics is the preferred tool for testing theoretical hypotheses. According to Popper, refutability is the testing of hypotheses with restrictive methods, and econometrics is one of these methods.
Some theoretical hypotheses are predictive: they predict causal links in the future that sometimes the data to test them are not available. It is in this type of case where the testing of assumptions is done by econometric forecasts as is the case with assumptions on the convergence/divergence of two aggregates that are tested by cointegration models based on forecasts
@ Zoubir, the present econometric methods on cointegration is always given cointegrated even when the results is not cointegrated. Example of such cases is the battle on energy consumption-Growth nexus; the controversy is much in the war on causality.
Actually, the only thing they have in common is that they both have something to do with Econometrics...
Hypothesis testing is using your model to examine whether the estimated parameters are in agreement with a certain economic theory, e.g. does income affect consumption?
Forecasting is using your model to predict future values of economic indices, e.g. will consumption in the following quarter will increase or decrease?
Hello
for those who do not see a link between econometric model forecasts and hypothesis testing, how to test such hypotheses:
The country's economic growth will not be able to remedy its external debt in the long term
Without isolation of infected individuals, the epidemic will reach city X in 3 days
@ Zoubir
Your two examples are forecasting exercises 'disguised' as hypothesis testing.
In statistical terms, hypothesis testing is always defined in terms of estimated parameters, whose values and significance may be tested.
One could had said instead that forecasting also involves the estimated parameters (e.g. confidence intervals are constructed that way). However, there are also non-parametric models...
@Zoubir, Pandelis, Gambo, then i ask what is the purpose of designing hypothesis if not forecasting. Again I stand to be corrected that before testing you have añswer yes or no. To me you forecasted before testing based on theoretical foundation.
@ Olatunji I already provided an example of hypothesis testing which its purpose is not forecasting above (i.e. income effect on consumption). In general, there are many cases where is not about forecasting at all, e.g. a case study for an microeconomic or macroeconomic event...
I am still surprised that you keep putting those terms even under the same umbrella...
@Pandelis, okay i understand where you are coming from. But still, i will cite my own example from macroeconomics. Malthusian theory is it a forecast or hypothesis.
Will your answer be different in the case of well known Philips Curve?
honourable professors
let's go back to the origins of estimation and significance tests used to test hypotheses deduced from theories (in management, not necessarily from a theory but also from empirical evidence). An estimate is indeed based on actual data (is not a forecast), but a significance test is an inference. An inference consisting in considering, hypothetically or following a test, that the variables follow a law, a priori the normal law. Inferring results obtained on a sample from a population is not the essence of the forecast? If you reject forecasting as a method of hypothesis testing, and I stress that today's issues are more complex than simply causal relationships on historical relationships, then we will also have to abandon econometrics, and if necessary significance tests and confidence intervals, as a tool for testing theoretical assumptions.
Respected Pandelis is right in his submission. But there is a point am trying to get with this question am asking. @Zoubir why do we rely on expectations during hypothesis and what does this expectation means?
@ Olatunji You seem to 'confuse' Economics with Econometrics, too.
The Malthusian theory and the Phillips curve are economic theories. That means that they assert/predict/describe that, under specific circumstances, economic growth and the inflation rate will have this kind of behavior/relationship with other variables...
On the other hand, hypothesis testing and forecasting are two (different) fields of Econometrics, for which a loose definition is 'statistical inference in economic issues'. Many Econometrics books (e.g. Aris Spanos' manuals) emphasize that testing and forecasting are actually two separate stages, when building an econometric model.
Therefore, 'inflation has an inverse relationship with unemployment', that is economic theory (Macroeconomics).
Running a regression of inflation rate on unemployment rate and testing the sign and significance of the unemployment rate coefficient, that is hypothesis testing.
Using this model (and unemployment rate projections) to predict future values of inflation, that is economic forecasting.
@ Pandelis, I really enjoy your comments so far. I understand the seperate connection between testing hypothesis and forecasting in econometrics and up to building a forecasting for Var family, GARCH faimily, ARIMA family among others.
In the two theories mentioned ealier were based on economic intuition tested with statistical inferences. I am trying to build in economic theory to a statistical forecasting system to solve resource curse issues.
One major query raised during the presentation in two conferences is based on the aforementioned statement which i need to unveil.
I do appreciate your analysis.
Thank you.
I do not see a clear demonstration of the 'resource curse issues' you state as the real reason for this post.
By the way, VAR, GARCH and ARIMA are all a-theoretical econometric models. i.e. they are not based on economic theory, and, again, I wonder how you relate them...
@ Pandelis, may be you wait when the work is out and you can see the relations.
@ Pandelis, so what do you think am searching for in the subject of discourse?
Hello to all of you
I will try to come back to some of the points discussed
The research hypothesis is an assumption derived from theoretical proposals. These proposals are in fact theoretical predictions, which sometimes leads to confusion between the hypothesis test and the predictions made by an econometric model. In this case, the sign of the parameter sign of the variable constructed to test the hypothesis and its statistical significance are interpreted to test the hypothesis. I just want to come back to a point that has already been made, that of inference, which is a generalization that is a prediction in the making.
I agree with Professor Pandelis' comments about theoretical models. These models are not actually used to test theoretical hypotheses. These are models used when theory cannot decide on a causal relationship, especially the number of delays of an explanatory variable. I would also add that non-parametric econometric models are part of a-theoretical models; their use is preferred when the form of the link between variables is not known and it is through the data that it will be necessary to search for it
Olatunji A Shobande I was referring to: "I am trying to build in economic theory to a statistical forecasting system to solve resource curse issues."
What do you mean by that? You want to build an economic theory that has good predictions? Or a new econometric method that does not need any theory?
About the models you mentioned, as Zoubir Faiçal said, they are a-theoretical. That means that they do not assume any relationship on advance, the theory may only serve to point out which (macro)economic time-series to use. Is that what you want to do? Ignore theory all together and only use Econometrics?
Then you must know that there also caveats in that case. E.g. You may find statistical results that cannot be explained by any human intuition... Therefore, they may be ... random, since two variables can be correlated 100% and be totally independent.
I will not ignored theories. I have made a construct to be used. It is a bit technical but i will achive it.
You will encounter strong opposition from purists in economics, by building a theoretical model or theory whose behavioural assumptions are not accepted by the economist community and where you include empirical data. Even the a-theoretical Var models are used in economics to test a theory that has not dealt more precisely with a component such as monetary economy shocks. Perhaps we will have to include our theoretical theory or model, whose predictions are not the result of deducing behavioural hypotheses in what is called in management: a managed paradigm
Olatunji A Shobande As long as you use Econometrics properly, then that's fine. E.g. you chose the model with the best statistical properties to make your conclusions (estimation, hypothesis testing and/or forecasting) and you make sure that the models you choose from have an economic theory to back them up. In the case you are using VAR, etc, then theory serves as to point out the variables and explain estimated coefficients.
This is actually standard economic research - or what standard economic research should be, at least. Good luck!
I have been using econometric and statistics for the 20years now. Were the work is going is a bit technical and am not using VAR is purely Dynamic Stochastic General Equilibrium (DSGE). I don't want to forecast alone.
MY Resource Curse Theory
The theory will supply components to general equilibrium model designed to address a specific economic question. The model is then calibrated by setting parameter values equal to average of economic ratios known not to have changed much over time.
One way to do this is to not rely exclusively on the theories developed within your particular discipline, but to think about how an issue might be informed by theories developed in other disciplines.
Sounds like in the end of the day you still have to decide which theory/discipline/variables matter the most. I smell Bayesian Model Averaging methods...
Thinking biophysical Economics : "Let human interact with nature"
The conceptual model upon which mainstream (neoclassical) economics bases its analyses consist of a circular flow of income between producers and consumers, mediated by means of markets. In this "perpetual motion" of interactions between firms that produce and households that consume, little or no accounting is given to the flow of energy and materials from the environment and back again, and little account given to human interactions that take place outside of market processes. Analyses by natural scientists (and others) find that the conventional model is simply not credible (see Hall and Klitgaard, 2019).
It takes a theory to beat a theory, though. And by theory, I mean a complete system of mathematical equations describing what kind of decisions the average persons (God help us) will make in terms of economic choices.
Don't forget that there is a huge literature describing e.g. how non-credible CAPM is, but it is still widely applied in one of the most challenging fields of Economics.
Anyway, this discussion created many and interesting story arcs but, on my behalf, it's now coming to an end.
Good luck!
@ Pandelis, it nice having a discussion of this magnitude with you. I am glad I exploit some knowledge from it. It is a discussion I will remembered for a long time.
> Is there any difference between testing hypothesis and econometric forecasting?
It seems possible to define them to be the same, arguing that any forecast must embody a hypothesis and that its success or failure is the test of that hypothesis. However, I do not think this is very helpful in practice because their focus is typically very different.
Hypothesis testing focuses on the hypothesis, making it as concrete and explicit as possible, and expressing it in a form designed for testing. The most rigorous disciples even invoke holy numbers such as 0.05 that sanctify the ritual.
Forecasting is a much vaguer and more general concept, encompassing guesses, induction and reasoning of many kinds that may not be expressed in numerical form, may not be entirely explicit and are typically not designed with testing in mind. A good poker player has high forecasting skill in her domain, but while we can admire the rate at which she accumulates chips -- econometric in the extreme -- it may be difficult to analyse how she does so in terms of specific hypotheses, even for her.
Hypothesis testing, when favourable conditions apply, provides an easy way to avoid being fooled by randomness. Unfortunately, there are many interesting and important problems that do not provide favourable conditions but for which forecasts must still be made.
So why not try to use econometric tools for them? Absent any strong theoretical basis you will need to throw in some induction and just assume that the future resembles the past in all the ways you need. This will of course sometimes be wrong, and the more complex and powerful the tool the more likely this becomes. However, econometric methods tend to lie somewhere in-between the straight-jacket of hypothesis testing and the free-for-all of wholly unstructured forecasts, and that can be a good place to be...?
- Badges
- Science topic
- Similar topics
- Philosophy
- Philosophy Of Science
- Reasoning