What is Forecasting?
Forecasting is the development of a forecast; in a narrow sense - special
The need for a forecast is due to the desire to knowfuture events, which is reliable, is impossible in principle, based on statistical (errors of current estimates), probabilistic (multivariance of consequences), empirical (methodological errors of models), philosophical (limited current knowledge) principles.
The accuracy of any forecast is due to:
- the volume of "true" (verified with a known error) initial data and the period of their collection;
- the volume of unverified source data and the period of their collection;
- properties of the forecasting object and the system of its interaction with the forecasting subject;
- forecasting methods and models.
With an increase in the set of factors affecting the accuracy of the forecast, it is practically replaced by routine calculations with a certain steady-state error.
Forecasts are divided (conditionally):
- by terms: short-term, medium-term, long-term, long-term;
- by scale: private, local, regional, sectoral, country, world (global);
- by responsibility (authorship): personal, at the level of the enterprise (organization), at the level of state bodies.
The main forecasting methods include:
- statistical methods;
- expert judgment (for example, the Delphi method);
- modeling methods, including simulation;
- intuitive (that is, performed without the use of technical means, impromptu, "in the mind" by a specialist who has experience in previously used scientific methods in this type of forecasts).
Statistical forecasting methods
Statistical forecasting methods - scientific andan academic discipline whose main objectives include the development, study and application of modern mathematical and statistical methods of forecasting based on objective data.
Development of theory and practiceprobabilistic-statistical modeling of expert forecasting methods; forecasting methods under risk conditions and combined forecasting methods using jointly economic-mathematical and econometric (both mathematical-statistical and expert) models.
The scientific basis of statistical forecasting methods is applied statistics and decision theory.
The simplest methods for reconstructing dependencies used for forecasting are based on a given time series, i.e., a function defined at a finite number of points on the time axis.
Assessing the accuracy of the forecast (in particular withusing confidence intervals) is a necessary part of the forecasting procedure. Typically, probabilistic-statistical models of dependence recovery are used, for example, they build the best forecast using the maximum likelihood method.
Parametric (usually based onnormal error models) and nonparametric estimates of the forecast accuracy and confidence limits for it (based on the Central Limit Theorem of Probability Theory). Heuristic techniques are also used that are not based on probabilistic-statistical theory, for example, the method of moving averages.
Multivariate regression, including the use of nonparametric estimates of the distribution density, is currently the main statistical forecasting tool.
Unrealistic assumption of normalityIt is not necessary to use measurement errors and deviations from the regression line (surface); however, to abandon the assumption of normality, it is necessary to rely on a different mathematical apparatus, based on the multidimensional Central Limit Theorem of probability theory, linearization technology and inheritance of convergence.
Forecasting applications
For forecasting using a time series, it is usuallyuse computer programs. This allows you to automate most of the operations when building a forecast, and also allows you to avoid errors associated with data entry and building models.
Such applications can be both local (foruse on a single computer) and Internet applications (available as a website, for example). Programs such as R, SPSS, Statistica, Forecast Pro, Forecast Expert should be distinguished as local applications.
What can be predicted?
- Weather
Errors in calculations of the future states of the atmosphere and other chaotic systems accumulate over time, so the weather forecast for a day ahead is much better than for a month.
However, the accuracygrowing gradually: modern five-day forecasts are just as goodas 40 years ago - one-day. A useful forecast can be made for nine to ten days. And the predictability limit for classical models, according to Alexander Chernokulsky, is two weeks.
All these models are built on the same principle.The weather is described by several basic equations, which are solved step by step by substituting observational data, and not in a general form, as taught in school - they simply cannot be solved that way.
In order not to end up in an awkward position, as Lorenz once did, the model is run 10-20 times, slightly changing the initial values - adding noise to consider different options.
- Magnetic storms
Scientists around the world have been working for 70 years toto find out the reasons for the abnormal heating of the solar corona. This process is associated with magnetic storms, which are still impossible to accurately predict.
Temperature of the solar corona - outer layerthe atmosphere of the Sun - is about 1 million degrees Celsius, and in some places it reaches almost 10 million. However, the lower atmosphere reaches only 5.5 thousand degrees.
As a result, the conclusion is: the further from the center of the Sun, the hotter it is, although inside it the opposite is true. The mechanism by which this heating of the corona works is still unclear.
Propagation of Alfvén waves Samarascientists are investigating using the equations of magnetic gas dynamics. Based on the results of the work, scientists will present systems of equations that mathematically accurately describe various parameters and models of heating the solar coronal plasma.
- Volcanic eruptions
Researchers at Stanford Universityanalyzed the location of olivine crystals that froze in lava after the eruption of Kilauea volcano. So scientists were able to find out the details of the processes taking place in the bowels of the earth - this information will help predict future eruptions.
Scientists explained that they tried to createan algorithm for predicting volcanic eruptions. However, many of the processes that might suggest this take place deep underground in lava tubes. After an eruption, any underground markers that might give clues to the explorers are destroyed in almost all cases.
So the researchers focused on studying olivine crystals that formed during a massive eruption in Hawaii more than half a century ago.
Thereafter, the Stanford researchersThe universities found a way to test computer models of magma flow, which they said could reveal more data about past eruptions and possibly help predict future ones.
- Fires
University fire laboratory researchBrigham Young's name in the United States provides a more accurate picture of where wildfires start and how they spread. Scientists are confident that any new data that will help control natural disasters will save the country's budget millions of dollars.
Studies have shown that the chemical compositionshrubbery is essential for how quickly they burn. The type of plant found near a fire can help predict how the fire will spread and how quickly it can spread to other plant species.
The experiment aims to improvefire forecasting models. Because they cost the US forestry service and government agencies billions of dollars annually, any research that can help make firefighting more effective is essential, the researchers noted.
- Climate change
Researchers from the Norwegian Business School in Oslocreated a mathematical model of climate change, according to which, after the cessation of all emissions, the temperature rise will continue for at least another 100 years.
The researchers used in their modelinformation about the climate from 1850 to the present day. Based on this, they predicted how global temperatures will change and how much sea level will rise before 2500.
As a result, it turned out that if the peak of emissionsgreenhouse gases will occur around 2030, and by 2100 will drop to zero, then by 2500 global temperatures will still be three degrees higher, and sea levels will be 2.5 m higher than in 1850. And this is the most favorable forecast.
Although some of the carbon dioxide from the air will be absorbed by biomass, soil and oceans, this will not stop global warming in any way. The point of no return was passed before 2020.
How can we improve our predictions?
In the future, data quality will improve thanks tospectroradiometers, radars and lidars (lasers) on new satellites. Advanced spacecraft are already capable of directing equipment if necessary.
Another promising area is measurements using ordinary smartphones equipped with all kinds of sensors and other consumer electronics.
There is another problem - with zooming outmodel and the growth of data volume, the complexity of calculations increases enormously. For example, weather forecasting uses some of the most powerful computers in the world.
They are expensive and their performance is higher.does not increase at the same rate: silicon microcircuits have almost nowhere to improve. In addition, modern meteorologists have a legacy of millions of lines of code, which makes the calculations difficult to optimize.
Read more
Physicists have created an analogue of a black hole and confirmed Hawking's theory. Where it leads?
Scientists have discovered the mythical particle of Odderon
The most mysterious natural phenomenon. Where does ball lightning come from and how is it dangerous?