Video Of Day

Breaking News

Triviality Of Parameter Doubtfulness Together With Touchstone Vibrations For Forecasting

In before articles, I discussed the notion of forecastability (link to previous article): is it possible to forecast the hereafter values of variables inward an economical model? This article volition create down an extended analysis of the simplest stock-flow consistent (SFC) model: model SIM. Based on what nosotros know virtually linear organisation theory, nosotros tin dismiss that 2 measure sources of dubiousness (measurement dissonance in addition to parameter uncertainty) are non forecasting challenges if nosotros assume that nosotros are working amongst the right economical model. Other sources of dubiousness introduce greater problems, in addition to volition hold upward discussed inward later on articles.

Introduction

In my commencement article on forecastability, I referred to model SIM. I conduct keep used this model extensively inward my discussions; if you lot desire to explicate a regulation virtually models, the most sensible technique is to apply it to the simplest possible model. For those of you lot unfamiliar amongst the model, more details are available inward this article, inward Godley in addition to Lavoie's textbook Monetary Economics (where I took the model from), or my majority of SFC models.

In interests of brevity, I volition non write out the model specification; at that spot is a surprisingly high number of equations for a rattling elementary model. The argue for the large number of equations is that it represents a total 3 sector model of the economy, amongst all accounting relations represented.

From a mathematical perspective, model SIM collapses to a elementary representation. There is a unmarried external (exogenous) variable - regime consumption (G), in addition to together amongst the previous period's coin stock decide all other variable's values. In particular, if nosotros aspect at a unmarried variable inward the model, the model to a greater extent than oft than non looks similar a commencement guild depression overstep filter system.

This article contains a broad number of assertions virtually such commencement guild depression overstep filter systems. I pass a few years of my life staring at this degree of systems, trying to consider how to derive useful results for a nonlinear variant. (Interested readers may wishing to consult "Input-Output Analysis of Feedback Loops With Saturation Nonlinearities" -- published nether the moniker "B.G. Romanchuk" -- to consider the slightly non-soporific results of that analysis.) So at that spot is a theoretical background behind these assertions, but I am skipping them inward interests of reader fourth dimension (and sanity).

The balance of this article discusses how nosotros generate forecasts for hereafter values for model SIM, inward the deterministic setting, equally good equally amongst unknown dissonance in addition to disturbances.

Suspension of Disbelief

In guild to become through this analysis, the reader needs to suspend their disbelief amongst regards to model SIM. Yes, model SIM is a terrible model of existent footing economies; that is a belongings it shares amongst almost all proposed mathematical economical models. However, inward guild to speak over the mathematical properties of model SIM, nosotros cannot flit betwixt discussing it, in addition to the properties of the existent footing or other hypothetical models inward a random fashion. Instead, nosotros demand to focus on model SIM alone, in addition to sympathise the mathematical properties of its forecasting power.

My hunch is that component of the problems inward academic economists' handling of mathematical models is a reliance on statistical tests to refuse models. They aspect at a proposed model, compare its forecasts to reality using some dark box statistical test, in addition to consider whether it passes. Although that mightiness overstep for "scientific" inward academic economist circles, I would complaint that this is non measure physical care for inward systems technology scientific discipline -- which is a patch that genuinely has some mathematical successes, dissimilar economics. If nosotros are discussing a mathematical model, nosotros demand to convey its dynamics seriously, in addition to those dynamics convey precedence over generic statistical tests. However, nosotros demand to believe inward our model dynamics. Since the full general expectation is that mathematical economical models fail, there is a disturbing lack of religious belief inward the model dynamics.

Forecasts With Parameter Uncertainty

If nosotros conduct keep all available information inward a deterministic model, nosotros tin dismiss patch forecast the hereafter values of  the model output past times but running the model. In guild for forecasting to hold upward non-trivial, nosotros demand some information to hold upward hidden from the forecaster. Since a formal specification of forecastability of models was non a concern to economists, the breakdown of public/private information tended to hold upward vague in addition to informal, in addition to would conduct keep to hold upward determined past times the reader.

If nosotros conduct keep the assumptions in addition to model dynamics of model SIM (as defined inward the references given earlier), I would combat that only existent source of dubiousness for forecasting is the menage sector consumption function. It is specified past times 2 "alpha" parameters (which are the propensity to eat of wealth in addition to income). Since I am non concerned amongst writing out the model equations inward detail, I volition label them hither equally $\alpha_1, \alpha_2$.

This beingness the only source of dubiousness may non fit intuition, but it appears to hold upward required if nosotros conduct keep the dynamics of model SIM.
  • If nosotros convey the green Definition of dry reason variable from dynamical systems theory, the only dry reason variable is the coin render (M) -- which also equals menage wealth. In systems theory, noesis of the dry reason is a typical source of uncertainty, which is why nosotros oft embed a Kalman filter into a controller dynamics to pivot downward the best guess of the state. However, the coin render is measured rattling accurately in addition to at rattling unopen to existent time, in addition to so the actual dubiousness virtually the organisation dry reason inward model SIM (if translated to the existent world) is negligible.
  • The occupation organisation sector acts inward an omniscient fashion inward model SIM, ever hiring precisely plenty workers so that it ever breaks even. In the existent world, at that spot would hold upward dubiousness virtually their powerfulness to do this. In model SIM, this omniscience is assumed past times the model dynamics, at that spot is literally no way to contain dubiousness inward the occupation organisation sector inside the model structure.
  • Government policy (consumption in addition to taxation rates) are exogenous, in addition to inward the existent world, cannot hold upward forecast perfectly. (Note that monetary policy does non be inward model SIM.) However, if nosotros interpret our model forecasts equally beingness conditional on an assumed path of hereafter regime policy, nosotros conduct keep eliminated that source of uncertainty. If regime policy changes relative to our assumption, our forecast would hold upward wrong. The key is that nosotros tin dismiss predict the forecast fault precisely based on the divergence of actual regime policy from what was assumed. That is the best that nosotros could promise for inward whatever mathematical model of the economy.
  • For simplicity, I volition assume that the taxation charge per unit of measurement is fixed for all time. We could suit the analysis to piece of work amongst a time-varying taxation rate, but nosotros destination upward amongst a linear time-varying system, which makes everything harder to piece of work with. Furthermore, the taxation charge per unit of measurement has to hold upward strictly positive (lying inward the interval (0,1)) inward guild for at that spot to hold upward a solution to the model.
The Definition of exact forecastability is specified that the forecasts are generated at a specific time, which nosotros announce T, in addition to nosotros conduct keep access to the next world information:
  • the fourth dimension serial of economical variables for the fourth dimension points 0,1,... T-1 (but patch non the consumption component parameters); and
  • the flat of regime consumption (G) for all time.
We tin dismiss in addition to then generate forecasts inward a straightforward fashion: commencement nosotros pivot downward the $\alpha$ parameters, in addition to then run a re-create of model SIM using those parameter values to generate the levels of variables for all fourth dimension greater or equal to T.

Getting the $\alpha_i$ estimates is an practice inward linear algebra. We announce the vector of the 2 parameters equally $alpha$, where $\alpha = (\alpha_1, \alpha_2)^T$, which agency it is a vector inward $R^{2x1}$. (The T superscript denotes the transpose of a vector/matrix.) At $t=0$, the flat of menage consumption $c$ is given by:
$$
c(0) = w_1(0) \alpha_1 + w_2(0) \alpha_2 = w(0) \alpha,  $$
where $w(t)$ is a 1x2 row vector that has entries equal to the menage disposable income in addition to previous period's wealth ($M$). (In this article, I announce the fourth dimension serial of consumption equally $c$, to distinguish it from a matrix, equally is green inward linear algebra. This does non fit the green economic science notation.)

Similarly,
$$
c(1) = w(1) \alpha. $$
We tin dismiss in addition to then stack upward the past times history of observed consumption in addition to $w$ vectors to larn the matrix equation:
$$
c = W \alpha,
$$
where $c$ is $T$ dimensional vector, in addition to $W$ is a matrix amongst $T$ rows in addition to 2 columns.

We in addition to then do some linear algebra.
$$
W^T c = W^T W \alpha,
$$
$$
\alpha = (W^T W)^{-1} W^Tc,
$$
assuming $(W^T W)^{-1}$ exists.

The weather for the matrix inverse to be are straightforward.
  • We volition demand $T \geq 2$ inward guild for a possibility for the matrix to hold upward non-singular. That is, nosotros demand at to the lowest degree equally many information points inward the dorsum history equally nosotros conduct keep parameters to estimate.
  • The choices for $G$ in addition to the inherited coin stock at $t=0$ are somewhat constrained. For example, they cannot hold upward equal to goose egg for all $t=0,1,...T-1$. This is annoying from the perspective of defining forecastability: it is possible to conduct inputs to homecoming an otherwise forecastable organisation non-forecastable. It is left equally an practice to the reader to demonstrate that the mensurate of the laid of inputs $G$ that homecoming the organisation non-forecastable is goose egg amongst honour to the laid of all possible inputs over the fourth dimension interval $0,...T-1$.
In patch English, so long equally nosotros conduct keep 2 historical information points, in addition to nosotros avoid pathological choices for inputs, nosotros tin dismiss in addition to then forecast the hereafter trajectory of model SIM perfectly (given the diverse assumptions).

However, exact forecastability is besides strong; nosotros intuitively know that nosotros volition non let on a model that tin dismiss do that amongst existent footing data. It is prophylactic to combat that anyone who feels that mathematical models are useful at all inward economic science accepts that nosotros desire "good enough" forecasts. In guild to justice on that criterion, nosotros demand to start introducing to a greater extent than sources of uncertainty. This article concludes amongst i source; later on articles volition speak over others.

Noisy Observations

Mathematically, the easiest source of dubiousness to add together is the possibility that dissonance corrupts our measurements. In this case, assume that nosotros cannot notice menage consumption $c$ directly, but only $\hat{c}(t) = c(t) + n(t)$, where $n(t)$ is the dissonance signal.

For an technology scientific discipline or physical system, the concept of dissonance is unremarkable: anyone who has worked inward a laboratory is used to adding inward an fault bar around measurements of physical variables. We but assume that $n(t)$ lives somewhere inside those fault bars.

In an economical model similar model SIM, nosotros know that accounting identities conduct keep to hold. So but having dissonance on i measuring is non enough; nosotros could dorsum out its truthful value past times applying accounting identities against non-corrupted variables. As a result, specifying dissonance inward these models has a hidden complexity which I desire to avoid. For our purposes here, assume that $n(t)$ reflects the dissonance that remains after nosotros conduct keep extracted equally much information from other variables.

From a forecasting perspective, nosotros tin dismiss but plough ahead in addition to apply the exact same algorithm to pivot downward the $\alpha$ vector. However, nosotros only hold off to converge to the best guess for $\alpha$ (under for certain assumptions, consider below) equally the length of our plumbing fixtures history lengthens. This is a fairly measure plumbing fixtures problem, in addition to its properties are good known.

In guild to do a proper mathematical analysis of this case, nosotros demand to start to pivot downward the properties of the dissonance signal $n$. There are 2 measure options (although I only holler upward seeing the commencement inward the economic science literature).
  1. We assume that $n(t)$ is generated past times a specific random (stochastic) process, using specific parameters for the probability distribution (mean, measure deviation, etc.).
  2. We but assume that $n(t)$ lies inward some measure infinite of fourth dimension signals amongst useful properties, such equally finite aAmplitude (infinity norm), finite release energy (2-norm) or finite powerfulness (finite 2-norm over finite intervals; this is non a measure mathematical space, but shows upward inward command systems).
The commencement selection -- using probability theory -- is the most complex mathematically. My impression is that some people equate mathematical complexity amongst modeling sophistication. However, I am inward the army camp that the stochastic approach is less sophisticated from a modelling standpoint (and obviously, the scholars producing stochastic command theory inquiry are probable to disagree amongst me).

The limitations of the probabilistic formalism are straightforward: the laid of all signals (with probability 1) generated past times a given stochastic physical care for is only a subset of the wider laid of signals inward the minute case, in addition to assuming a specific laid of stochastic parameters to specify the dissonance signal requires a peachy bargain of certainty virtually an allegedly unknown dissonance signal. 

In technology scientific discipline systems, what nosotros think of equally "noise" is oft non rattling random. Anyone who did electronics laboratory piece of work at McGill University had the joy of having their circuits oscillate equally a upshot of the broadcasting from the nearby transmitting tower on Mount Royal. Although the signal looks similar a brownian motion, careful analysis volition tell you, that yep, that is "Stairway to Heaven" beingness frequency modulated at 97.7 MHz. Importantly, some the "noise" is materials that nosotros genuinely conduct keep to worry about. The air conditioning electrical flow coming into the powerfulness render is non a pure sinusoid, in addition to whatever circuit that does non conduct keep a top tier powerfulness render volition tend conduct keep to signals at around sixty Hz (in my component of the grid), in addition to at the diverse harmonics (120 Hz, 180 Hz). Those frequencies are unopen plenty to the high frequency harmonics of physical systems that nosotros demand to pay some attending to that "noise" inward our analysis. This tin dismiss hold upward done past times shaping the random physical care for used inward analysis, but doing so is a PITA relative to the deterministic frequency domain dissonance shaping approach.

We conduct keep far less information available when working amongst economical models, in addition to so the defects that exhibit upward inward technology scientific discipline may hold upward less pressing there. However, the stochastic formalism results inward far to a greater extent than opaque mathematics, requires a large amount of certainty virtually the stochastic process, in addition to in all likelihood requires throwing out known information virtually disturbances inward guild to proceed analysis tractable.

We tin dismiss consider the advantages of the non-stochastic approach inward this example. We but demand to enquire a elementary question: what volition crusade the guess for $\alpha$ to diverge from the truthful value? One obvious candidate is that nosotros let on a dissonance signal that plants the observed $\hat c$ to equal the truthful output generated past times some other incarnation of model SIM amongst a different $\alpha$ parameters (call it $\tilde \alpha$. Under this construction for $n(t)$, nosotros volition ever converge to ${\tilde \alpha}$ equally our best guess for the parameters.

We tin dismiss in addition to then consider the Achilles Heel of the construction: the dissonance signal has to hold upward an extremely persistent variable, equally the dissonance signal is the difference betwixt the outputs of 2 distinct linear systems. In the limit, this dissonance signal would prevarication exterior the infinite of signals of 2-norms, in addition to is hence non a valid dissonance signal past times the deterministic formalism.

Admittedly, it would convey a flake of mathematics to fill upward inward the details. But I would assert that nosotros would hold upward comfortable amongst our forecasts for model SIM if nosotros tin dismiss assure ourselves that our measurements are non beingness systematically distorted past times persistent measuring errors inward our plumbing fixtures information set. And if that is the case, the occupation is non our economical model, rather that nosotros conduct keep made a substitution fault inward how nosotros are measuring the variables inward our system, in addition to at that spot is no way that any economical model tin dismiss overcome such gross incompetence. 

One could in all likelihood come upward up amongst a similar analysis using the stochastic formalism. The number is that nosotros would hold upward stuck amongst especial models for the random process, in addition to the logic would hold upward buried inward a dog's breakfast of stochastic calculus.

Concluding Remarks

We conduct keep covered the 2 simplest ways of implementing dubiousness inward an economical model -- parameter uncertainty, in addition to measuring noise. Follow upward articles volition speak over agency to create to a greater extent than difficulties for economical forecasting. Another subject of involvement is the news of the criterion for rejecting models. As i mightiness guess, the writer is skeptical virtually the blind application of statistical tests.

Appendix: Data Measurement Lags

In economical models, i measure occupation is the reality that information is measured amongst a lag. We would correspond this past times having the measured signal hold upward the truthful signal passed through a lag filter. This transformation does non run across the green Definition of beingness noise, since it is a deterministic transformation of the fourth dimension series.

In command engineering, lags are treated amongst respect. Anyone familiar amongst the physical care for of showering volition sympathise why. It takes fourth dimension for a alter inward the temperature command dial to exhibit upward inward the observed temperature of the water. If nosotros are besides impatient, nosotros tin dismiss destination upward oscillating betwixt besides hot in addition to besides cold. After a few year's experience amongst this phenomenon, most people figure out that they demand to react amongst a lag to temperature changes, giving fourth dimension to for previous adjustments to exhibit up. In such an environment, it is impossible to command the temperature of the H2O at a high frequency (if for some crazy argue you lot wanted your shower temperature to accurately rail a pre-determined trajectory).

As long equally nosotros know what the measuring lag is, nosotros tin dismiss adjust out plumbing fixtures algorithm to convey it into account, in addition to nosotros volition in i lawsuit once again destination upward amongst the best guess of $\alpha$. For model SIM, nosotros volition in addition to then conduct keep almost no difficulties amongst forecasting afterward (unless our estimates of regime spending in addition to the coin render are lagged, which is somewhat implausible).

Other models that are to a greater extent than heavily theme upon dry reason information would conduct keep greater difficulties amongst forecasting inward the presence of measuring lags. Our forecasts would hold upward conditional on our estimates of the electrical flow values of variables, which nosotros would base of operations on other real-time information equally good equally the lagged information. Whether the organisation is forecastable or non would depend on our powerfulness to infer the electrical flow values of variables -- which would depend upon the model specification. 

(c) Brian Romanchuk 2018

No comments