3
Term | Meaning |
---|---|
3 Period Moving | The 3-month moving average percent change used for the regression. See also: Calculation Adjustment |
3 Period Moving Period over Period | The average of the last 3 periods. See also: Calculation Adjustment |
3 Period Year over Year | The percent change of the 3-period moving average from the prior year’s three period moving average. See also: Calculation Adjustment |
A
Term | Meaning |
---|---|
Accuracy | This tab allows you to conduct back tests to measure forecast quality through analysis of the model’s out-of-sample performance. By selecting an "analysis as of date", you can simulate a back test in the app at your chosen point-in-time. The app will assume the availability of all data from all periods up to the chosen point-in-time. See also: Single Period Error and Aggregate Error |
Accuracy of Insights | A measure of how accurately the independent variables explain the dependent variable. It ranges from 0-100%, where 100% means the model has no bias, in which predictions do not systemically over- or under-estimate. See also: Model Score |
Accuracy of Predictions | A measure of how accurately the model produces the forecast. It ranges from 0 to 100%, where 100% means that we can rely on the predictions of the model with confidence. See also: Model Score |
Aggregate Error | The percentage difference between the sum of the forecasted value in all periods between the "analysis of date" and the stated period and the sum of the actual values in the same periods. See also: Accuracy |
Aggregation | The rule that is used when transforming data to lower level frequencies. e.g. Monthly volume data is aggregated to quarters by summation. A monthly index may be aggregated to quarters by using the average across the three months for each quarter, or by using the last month in the quarter. |
Analysis As of Date | A cutoff point that allows users test how well their model would have performed using only information available up to that specific date. |
Annualized Model Error Trend (Intercept) | The parameter in a multiple linear regression model that gives the expected (error) value of the dependent variable when all the independent variables are equal to zero (annualized). See also: Component Contribution |
Annualized Model Error Trend (Slope) | This is the line of best fit, which best expresses the relationship between residuals. See also: Component Contribution |
Average Aggregate Error | This value displays how consistently accurate was the model for a certain cumulative period, e.g. 3 months out. This value is calculated by finding the error rate for an interval period (1-month + 2-month + 3-month forecast) and comparing that result to actuals for the same period (1-month + 2-month + 3-month actuals) to calculate the error rate. The calculation is then repeated back to the regression start date. Finally, the results are averaged. |
Average Single Period Error | This value displays how consistently accurate the model was for a certain period, e.g. how accurately does the model predict 4 months out? It is calculated by looking at the forecasts for a single period (e.g. month 4) and then comparing it to the actuals for the same period, finding the error rate. The calculation is then repeated all the way back to the regression start date. Finally, the results are averaged. |
B
Term | Meaning |
---|---|
Breusch Gallery AR 4 | This test is used to check for autocorrelation in the residuals from a regression analysis. An AR 4 indicates that the test is specifically checking for autocorrelation up to lag 4. A significant result suggests that the model may need adjustments to account for this autocorrelation. |
C
Term | Meaning |
---|---|
Calculation | Represents the transformation of data series for comparison. e.g. None, 3-period moving average, % change over time (period over period, year over year). |
Calculation Adjustment | Apply various calculations to the indicators used in the workbench. See also: 3 Period Moving, 3 Period Moving Period over Period, 3 Period Year over Year, Seasonal Adjustment, Knit Series, None, and Period over Period. |
Component Contribution | The component contribution decomposes the forecast into the individual contributions of each underlying indicator across both historical and projected periods. This stacked bar visualization displays how each variable positively or negatively impacts the overall prediction at monthly, quarterly, semiannual, or annual frequencies See also: Annualized Model Error Trend (Intercept), Annualized Model Error Trend (Slope), Diagnostics, Durbin Watson, In Sample Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Raw vs. 3 Period Year over Year, Relative Importance, Root Mean Square Error (RMSE) |
Consistency Exclusion | The exclusion of a time period from data when analyzing or modeling forecasts. |
D
Term | Meaning |
---|---|
Diagnostics | Various statistical tests and factors that determine the statistical integrity of the model. This includes residuals, statistical tests, and correlation matrix. See also: Residuals |
Directional Symmetry 12 Periods | A statistical measure that displays the percent of the time the model output and the actuals were moving in the same direction over the last 12 periods. |
Durbin Watson | A test for whether a residual series is related to its immediately preceding values, i.e. a test for first order autocorrelation. See also: Diagnostics |
E
Term | Meaning |
---|---|
Explained Sum of Terms (ESS) | The portion of the historical variation in the dependent variable that is explained by a least squares regression model. |
F
Term | Meaning |
---|---|
Forecast | The process of predicting the future values of a dependent variable using a combination of independent variables. See also: Period over Period, Raw, Three Period Year over Year, and Year over Year. |
Frequency | The time interval in which data is collected. |
H
Term | Meaning |
---|---|
Heteroskedasticity (Or Heteroscedasticity) | A condition of a data series in which the variance is not constant across varying time intervals or independent predictors. |
I
Term | Meaning |
---|---|
Indicator | A time-value data series. |
In Sample Mean Absolute Error (MAE) | The measure of the average absolute difference between actual and predicted values, expressed in the same units as the data. One of the simplest and most widely used measures of forecast accuracy. See also: Diagnostics |
K
Term | Meaning |
---|---|
Knit Series | Remove outliers from a series. |
M
Term | Meaning |
---|---|
Mean Absolute Percentage Error (MAPE) | A common measure for evaluating the accuracy of a forecast or predictive model. It expresses the prediction error as a percentage, making it easy to interpret across different scales. See also: Diagnostics |
Model | Allows users to create and/or access forecasts that for a project. The model page allows users to change and tune the mode. To edit the model, it must be in an "in progress" state, ‘active’ models cannot be edited. |
Model Efficiency | A measure of how stable the coefficients are as we add new data to the model. It ranges from 0-100%, where 100% means the model’s coefficients are very stable. See also: Model Score |
Model Performance | A measure of the overall performance. It is comprised of MAPE, Residual Trend, and Predictive R-squared. See also: Model Score |
Model Quality Score | A comprehensive quality score based on criteria such as error metrics and other statistical evaluations that gauge how well a model predicts the data based on its parameters. See also: Model Score |
Model Score | A comprehensive score for your model that includes Model Bias, Model Efficiency, Forecast Accuracy, Overall Model Performance, and Penalties. It ranges from 0 to 1 (100%), where 1 is the best score. The score can be negative if there is a penalty. |
Multicollinearity | A condition where two or more of the explanatory variables used in a regression model are correlated to one another. |
Multiple Regression | A statistical technique that uses least-squares methodology and more than one explanatory variable to predict the outcome of a response variable. |
N
Term | Meaning |
---|---|
None | The raw values of the indicators are used. See also: Calculation Adjustment |
O
Term | Meaning |
---|---|
Outliers | Data points that do not fit in with the pattern of the other observations in a dataset. These are values that appear significantly higher, lower, or otherwise different from what you'd normally expect based on the rest of the data. There is no universal measurement to declare outliers in a series. |
Overfit | A condition where a model becomes too closely tailored to explain specific historical data at the expense of its ability to make accurate predictions on new, unseen data. This typically occurs when using too many variables relative to the amount of available data, or when including explanatory variables that show misleading relationships with the target variable during the sample period but don't represent genuine underlying patterns. An overfit model performs well on historical data but poorly when predicting future outcomes. |
P
Term | Meaning |
---|---|
Penalty | A mechanism that adjusts model scoring by penalizing undesirable characteristics such as unexpected coefficient relationships. Penalties provide a way to incorporate additional criteria beyond basic statistical fit, ensuring models are evaluated not just on how well they explain historical data, but also on whether they exhibit economically sensible behavior and maintain robust predictive power. |
Period over Period | Represents the percent change from the prior period. See also: Forecast |
Primary Indicator (Dependent Variable) | The main independent variable of interest. The key factor being studied to understand its relationship with the independent variables, whether in a workbench or model. |
R
Term | Meaning |
---|---|
Raw | Represents the untransformed values of units from period to period. See also: Forecast |
R (Correlation Coefficient) | A standardized measure, bounded between -1 and 1, that quantifies the strength and direction of the linear relationship between two variables. Also known as Pearson's r. Values closer to +1 indicate a stronger positive linear relationship, while values closer to -1 indicate a stronger negative (inverse) linear relationship. A measure near zero suggests no meaningful linear relationship exists between the variables. |
Regression Start Date | The earliest date from which data is included in the regression model estimation. This date determines the beginning of the sample period used to calculate the model's coefficients and statistical relationships. |
Relative Importance | A measure quantifying how much each indicator contributes to explaining the historical variance in the dependent variable over the time period of the model. This metric helps identify which variables have the greatest explanatory power in the historical model performance. |
Residuals | The differences between the actual values of the dependent variable and the values that the model estimated for them (as in, the parts of the dependent variable that the model could not explain). See also: Diagnostics |
Root Cause Analysis Chart | This indicates whether the average error is shown in the legend of the chart, providing context regarding the error rates across the analyzed periods, making it easier for users to understand the model's forecasting accuracy over time. This setting controls whether the displayed results are shown as percentages, which allows for easier interpretation of how individual elements contribute to the overall performance and variance |
Root Cause Calculation | An analysis of the residuals to determine underlying causes for any deviations or unexpected values in the forecast compared to actual data, helping to identify which factors might be influencing model performance adversely. |
Root Mean Square Error (RMSE) | The standard deviation of the residuals (prediction errors). RMSE measures the average magnitude of the error. See also: Accuracy |
S
Term | Meaning |
---|---|
Seasonal Adjustment | The process of removing period-to-period fluctuations that occur consistently every year from a data series |
Secondary Indicator (Independent Variable) | The explanatory variables which will be compared to the primary. In modeling framework, may be called exogenous variables. |
Segmented Multiple Regression | An extension of least-squares multiple regression where the data is partitioned into distinct intervals, with separate regression relationships estimated for each segment. This approach allows the model to capture different coefficient relationships across time periods or conditions, rather than assuming a single linear relationship holds throughout the entire dataset |
Single Period Error | The percentage difference between the forecasted value and the actual value in the stated period. See also: Accuracy |
Snapshots | The different versions of models. Can be used to retrieve various iterations of the same model and/or track their model’s progress. |
Start Date | The user-defined beginning date for analyzing statistical relationships within a workbench of indicators or model. This parameter allows you to specify when the analysis period begins, enabling you to examine variable correlations, model performance, and other statistical metrics from that point forward while excluding earlier historical data from the evaluation. |
T
Term | Meaning |
---|---|
Target | Creates a comparison point for the dependent variable forecast. Can be an internal forecast, or some other baseline, that the user can use to compare with Foresight generated forecast model. |
Trim (Fixed or Rolling) | The user-defined parameter that determines the length and endpoint of a forecast. A rolling trim extends the forecast for a specified number of periods from the end of the dependent variable's historical data (e.g., 6 months, 2 quarters). A fixed trim sets a specific end date that the forecast will not extend beyond, regardless of when the historical data ends. |
U
Term | Meaning |
---|---|
Univariate Forecast | Predict future values based on historical data. |
W
Term | Meaning |
---|---|
Waterfall | A visualization that shows how individual components contribute to a cumulative total by displaying each indicator's positive or negative impact in raw numbers as sequential steps. Starting from a baseline, each bar represents an indicator's numerical contribution, with the bars flowing from left to right to build up to (or subtract from) the final forecast value. |
Workbenches | A dashboard-like feature that allows users to assess relationships between a primary indicator and secondary indicator(s). |
Y
Term | Meaning |
---|---|
Year over Year | Represents the percent change from the same period in the prior year. See also: Forecast |