Glossary

A-D | E-L | M-R | S-Z

A-D

accuracy - describes the centrality of data groupings around a true value. (see precision)

advanced mathematical techniques - recently developed methods that are a major advance in data processing and statistical processing.

associated series - a data series that has been found to share mutual information with the target series. Associated series are used to build the multiple forecasting models which in turn are used to form the final forecast for the target series. (see parallel series)

Bayesian Analysis - a technique for optimizing the signal to noise decomposition such that a forecasting model is proportionate to the input data. (see cross validation)

bias - the effect of external influences (such as prior experience or expertise) on an outcome. (i.e. pertaining to a crime, a victim might have a different testimonial than a juror because of emotional bias)

committee of networks - averaging multiple models, each one of which is over-fitted, but where the errors associated with fitted noise self-cancel

Company of Chartered Surveyors - a non-profit organization that supports member education and the advancement of property research.

cross validation - the alternative signal to noise decompostion method to Bayesian Analysis. Inferior to Bayesian Analysis because it requires masses of data.

data-led - a term used to differentiate forecasts generated exclusively from data from forecasts generated using a theory or hypothesis basis. (...more)

data-led forecast - a forecast generated from data-led techniques. This method of forecasting eliminates bias that is intrinsic to a hypothesis-based model. In general, data-led forecasts are much less subjective than other methods.

decision-making - the process of acting upon the best information available to determine a single course of action for future direction.

deterministic - having a fixed outcome bounded by strict principles or methods. (see heuristic)

E-L

econometric - any method that combines economics with mathematical techniques. In many cases this method requires a subjective hypothesis around which the mathematical model is constructed.

experienced appraisal - an evaluation or review done by a qualified individual in the specific field of interest. For the context of this site, this would refer to property appraisal.

explanation/ prediction dilemma - a coined phrase that explains that prediction models based around well-understood theories provide good expanations for the model, but do not necessarily provide good predicitons. Alternatively, sound mathematically-based, non-linear prediction methods do not provide good explanations for associated predictions. This implies that if one wants good predictions one must learn to live with black boxes and an understanding of the meaning of probability.

forecast - the act or product of an act of projecting the most probable future values of a data series.

forecasting - the act of using some defined method of projecting the future value(s) of a data series

forecasting model - the underlying tool that allows a forecast to be created. Mathematically-based in data-led forecasting.

heuristic - referring to manual methods; generally accepted practice sometimes referring to a trial and error process. (see deterministic)

inference - to draw a conclusion that is not explicit to the subject matter

indices - plural for index.

index - a collection of data that has been collated and presented in such a form as to be recognized as a measure of performance

IPD Monthly Pages - a databank of property indices collected and calculated by Investment Property Databank (IPD). The IPD Monthly Pages are recognized by property professionals in the commercial property industry throughout the U.K.

Information Theory - a specific field of study pioneered by Claude Shannon. (links)(see mutual information)

linear - having or demonstrating a directly proportional output for a change in input.

M-R

masking - the effect of covering up or hiding the true nature or representation of data in a series. This is most often an unintentional occurrence and is a significant problem in any data series.

mechanical forecasting - another name for technical forecasting

multi-layer perceptron (MLP) - a neural network configuration that relies strictly on feed-forward inputs and that has representational capabilities characterized by either threshold or sigmoidal activation functions. (see Radial Basis Function)

mutual information - in the context of Information Theory, describes the situation where one data set can be mathematically proven to have some effect on another series. (i.e. When OPEC announces a cap on oil production, it is widely recognized that the price of oil can be expected to increase. In this example, the data series represented by oil production can be said to share mutual information with oil prices although it is not proven mathematically in this example) (...more)

noise - a term used to describe the random nature of data to stray from its true value. This can be analagous to the hissing sound emanating from a radio when the signal is adversely affected by atmospheric conditions or extreme distances.

non-linear - describes the state where an outcome does not give a directly proportional output for a change in input. Frequently used to describe the mathematical characteristics of a data series. Using a linear forecasting model to forecast linear data gives highly restrictive results. Some treatment of a non-linear data series is required to make a linear forecasting model function. Best practices require the application of non-linear methods to data which will then accommodate both linear and non-linear data regardless.

Occam's Razor - which can be paraphrased by saying that the simplest model capable of representing data, is usually the best. That general advice pre-supposes knowledge of signal and noise components of the time series and the appropriate level of complexity for a network function.

parallel series - a data series of either an economic or financial nature that will be considered as a potential candidate for a forecasting model. If mutual information theory identifies a parallel series as sharing a high degree of mutual information with the target series, the parallel series is likely to be used in a forecasting model. At this stage it becomes an associated series.

precision - a description of the amount of deviation from a specific value that is not necessarily the true or actual value. (see graphic) (see accuracy).

qualitative - a description of an event or occurrence that conveys the sentiment and feeling without using specific measures. (i.e. '...a long walk...')

quantitative - a way to describe an event or occurrence using specific values and measures typically avoiding sentiment and/or feeling. (i.e. '...a 3.4 km walk...')

radial basis function (RBF) - the preferred TFL method of forecasting data. Pioneered by TFL principals in their application to neural computing, RBFs are a specialized neural computing technique. Contrary to multi-layer perceptrons (MLP) they are particularly well suited to forecasting relatively stable times series such as property indices. Outperforms MLP in speed and stability when used with Bayesian Analysis.

related financial instruments - financial based data series or indices

Royal Institute of Chartered Surveyors (RICS) - A self regulated professional society governing the educational requirements and professional practice of property surveyors in the U.K.

S-Z

series - a collection of related data points along a time line. Normally describes a historical trend and is the basis for a data-led forecast. Usually referred to as a data series or data set.

shock - any completely unforeseen event that causes a dramatic change to a data series. (examples of this would be the World Trade Centre disaster in 2001 which sent ripples throughout the worldwide economic community, or the collapse of banks across the world in 2008 caused by the 'toxic home loans' scandal in USA)

signal - used in context of a data series, this term pertains to the the 'true' value each data point should take in the set. For many reasons, the true signal in data can be hidden by the associated noise. Reasons for this phenomenon could be as simple as inaccurate data collection or recording methods, or as complex as masking resulting from unclear differentation of data sources or classes. This makes forecasting the data more difficult and sometimes impossible.

subjective - open to different interpretations based on prior experience or expertise. Having no definitive or specific interpretation. Having a broad range of answers depending on the basis for the response.

subjectivity of expert opinion - see subjective used in the context of expert opinions. In this usage, the range of 'answers' may be narrower than for the general population.

target series - the specific data series of interest for which the forecast is to be created

technical forecasting - a specialized area of forecasting that forms its basis heavily around data-led techniques. This type of forecasting method tends to minimize SOFT inputs and opinion. (see also mechanical forecasting)

technique - a specific method or series of steps that is repeatable and can be differentiated from other possibly similar methods

unique - statistically independent from other series (when used in the context of data-led forecasting)

Valuation Office Agency (VOA) - the U.K. government's agency whose mandate is to provide property valuations to other government departments for tax, financial planning and other property related purposes