Univariate Forecasting Models Research Articles

COVID-19 İnsidans Oranlarını Modelleme ve Öngörümleme: Fransa'da Gözetim Başlangıcından Beri Akut Solunum Enfeksiyonları (ARI) Zaman Serisi Analizi

Objective: This study aims to address the challenges of planning and managing the trajectory of the COVID-19 pandemic by evaluating the predictive abilities of three distinct forecasting models. The primary focus is on the ATA univariate forecasting method, ARIMA (AutoRegressive Integrated Moving Average), and ETS (Error-Trend-Seasonality) models. These models are applied to a meticulously collected dataset comprising Acute Respiratory Infections (ARI) incidence rates in France, systematically collected since the initiation of surveillance. Methods: The purpose of the study was to conduct a comprehensive evaluation of forecasting models using the selected dataset to achieve its objective. The focus was on comparing the accuracy and performance of ATA univariate forecasting, ARIMA, and ETS models in predicting COVID-19 incidence rates. Additionally, the study incorporated a combination approach proven to be effective in enhancing forecasting performance. Results: According to the results obtained regarding forecast performance, the univariate models indicate that the ATA method exhibits the highest performance, while observations reveal that combinations of ATA and ARIMA methods enhance forecast accuracy. Conclusions: In summary, the most accurate approach for forecasting future Covid-19 incidence rates, specifically those derived from Acute Respiratory Infections (ARI), has been a combination of the high-accuracy methods ATA and ARIMA. These findings enhance our understanding of the trajectory of the pandemic, providing a foundation for strategic planning and effective management.

On the update frequency of univariate forecasting models

In univariate time series forecasting, models are typically updated at every single review period. This practice, which includes specifying the optimal form of the model and estimating its parameters, theoretically allows the models to exploit new information and to respond quickly to possible structural breaks. We argue that such updates may be irrelevant in practice, also unnecessarily increasing computational cost and forecast instability. Using two large data sets of monthly and daily series as well as an indicative family of conventional time series models, we investigate several model updating scenarios, ranging from complete model form specification and parameter estimation at every review period to no updating at all. We find that intermediate updating scenarios, including the re-estimation of specific parameters but not necessarily the specification of the model form, can result in similar or even better accuracy with significantly lower computational cost. We also show that similar conclusions hold true for popular machine learning methods, as well as for setups where different approaches are utilized for training the models or accelerating their specification and estimation. We discuss the implications of our findings for manufacturers, suppliers, and retailers and propose avenues for future advances in the area of model frequency updating.

Predictive model of acute kidney injury after spontaneous intracerebral hemorrhage: A multicenter retrospective study.

Acute kidney injury is a common comorbidity in patients with intracerebral hemorrhage. Although there are predictive models to determine risk of AKI in patients in critical care or post-surgical scenarios or in general medical floors, there are no models that specifically determine the risk of AKI in patients with ICH. Clinical features and laboratory tests were selected by previous studies and LASSO (least absolute shrinkage and selection operator) regression. We used multivariable logistic regression with a bidirectional stepwise method to construct ICH-AKIM (intracerebral hemorrhage-associated acute kidney injury model). The accuracy of ICH-AKIM was measured by the area under the receiver operating characteristic curve. The outcome was AKI development during hospitalization, defined as KDIGO (Kidney Disease: Improving Global Outcomes) Guidelines. From four independent medical centers, a total of 9649 patients with ICH were available. Overall, five clinical features (sex, systolic blood pressure, diabetes, Glasgow coma scale, mannitol infusion) and four laboratory tests at admission (serum creatinine, albumin, uric acid, neutrophils-to-lymphocyte ratio) were predictive factors and were included in the ICH-AKIM construction. The AUC of ICH-AKIM in the derivation, internal validation, and three external validation cohorts were 0.815, 0.816, 0.776, 0.780, and 0.821, respectively. Compared to the univariate forecast and pre-existing AKI models, ICH-AKIM led to significant improvements in discrimination and reclassification for predicting the incidence of AKI in all cohorts. An online interface of ICH-AKIM is freely available for use. ICH-AKIM exhibited good discriminative capabilities for the prediction of AKI after ICH and outperforms existing predictive models.

Predicting Thermoelectric Power Plants Diesel/Heavy Fuel Oil Engine Fuel Consumption Using Univariate Forecasting and XGBoost Machine Learning Models

Monitoring and controlling thermoelectric power plants (TPPs) operational parameters have become essential to ensure system reliability, especially in emergencies. Due to system complexity, operating parameters control is often performed based on technical know-how and simplified analytical models that can result in limited observations. An alternative to this task is using time series forecasting methods that seek to generalize system characteristics based on past information. However, the analysis of these techniques on large diesel/HFO engines used in Brazilian power plants under the dispatch regime has not yet been well-explored. Therefore, given the complex characteristics of engine fuel consumption during power generation, this work aimed to investigate patterns generalization abilities when linear and nonlinear univariate forecasting models are used on a representative database related to an engine-driven generator used in a TPP located in Pernambuco, Brazil. Fuel consumption predictions based on artificial neural networks were directly compared to XGBoost regressor adaptation to perform this task as an alternative with lower computational cost. AR and ARIMA linear models were applied as a benchmark, and the PSO optimizer was used as an alternative during model adjustment. In summary, it was possible to observe that AR and ARIMA-PSO had similar performances in operations and lower error distributions during full-load power output with normal error frequency distribution of −0.03 ± 3.55 and 0.03 ± 3.78 kg/h, respectively. Despite their similarities, ARIMA-PSO achieved better adherence in capturing load adjustment periods. On the other hand, the nonlinear approaches NAR and XGBoost showed significantly better performance, achieving mean absolute error reductions of 42.37% and 30.30%, respectively, when compared with the best linear model. XGBoost modeling was 8.7 times computationally faster than NAR during training. The nonlinear models were better at capturing disturbances related to fuel consumption ramp, shut-down, and sudden fluctuations steps, despite being inferior in forecasting at full-load, especially XGBoost due to its high sensitivity with slight fuel consumption variations.

Time series modeling and forecasting of vegetable production in United Arab Emirates

Crop forecasting is a formidable challenge. The national and state governments need such predictions before harvesting for various policy decisions relating to storage, distribution, pricing, marketing, importexport etc. In this paper, univariate forecasting models such as random walk, linear trend, quadratic trend, exponential trend, S-curve trend, simple exponential smoothing, Holt's linear exponential smoothing and Autoregressive Integrated Moving Average (ARIMA) models are used to predict vegetable production in the United Arab Emirates. For empirical analysis, a set of 9 different vegetable groups have been considered, contingent upon availability of required data. Annual data from 1974–75 to 2018–19 was used to forecast the next five years since 2019. Suitable models were selected based on the lowest RMSE and minimum of AIC criterion. Model diagnostic checking was done through Runs above and below the median, Runs up and down and Ljung-Box tests on ACF and PACF of residual terms. For onions and green shallots linear trend model was selected as the best fit, whereas simple exponential smoothing model was most suitable in cauliflowers and broccoli, pumpkins, squash and gourds and spinach. The optimum model obtained for forecasting carrots and turnips was Holt's linear exponential smoothing model and ARIMA model was the best fit for the rest of vegetable groups.

Ensembles of localised models for time series forecasting

With large quantities of data typically available nowadays, forecasting models that are trained across sets of time series, known as Global Forecasting Models (GFM), are regularly outperforming traditional univariate forecasting models that work on isolated series. As GFMs usually share the same set of parameters across all time series, they often have the problem of not being localised enough to a particular series, especially in situations where datasets are heterogeneous. We study how ensembling techniques can be used with generic GFMs and univariate models to solve this issue. Our work systematises and compares relevant current approaches, namely clustering series and training separate submodels per cluster, the so-called ensemble of specialists approach, and building heterogeneous ensembles of global and local models. We fill some gaps in the existing GFM localisation approaches, in particular by incorporating varied clustering techniques such as feature-based clustering, distance-based clustering and random clustering, and generalise them to use different underlying GFM model types. We then propose a new methodology of clustered ensembles where we train multiple GFMs on different clusters of series, obtained by changing the number of clusters and cluster seeds. Using Feed-forward Neural Networks, Recurrent Neural Networks, and Pooled Regression models as the underlying GFMs, in our evaluation on eight publicly available datasets, the proposed models are able to achieve significantly higher accuracy than baseline GFM models and univariate forecasting methods.

Forecast and error analysis of vegetable production in Haryana by various modeling techniques

Crop forecasting is a formidable challenge for every nation. The Government of India has developed a number of forecasting systems. The national and state governments need such pre-harvest forecasts for various policy decisions on storage, distribution, pricing, marketing, import-export and many more. In this paper, univariate forecasting models such as random walk, random walk with drift, moving average, simple exponential smoothing and Autoregressive Integrated Moving Average (ARIMA) models are considered and analyzed for their efficiency for forecasting vegetable production in the Haryana state. The State annual data on vegetable production were divided into the training data set from 1966-67 to 2013-14 and the test data set from 2014-15 to 2018-19. Suitable models were selected on the basis of error analysis on the training data and a percent error deviation test on the test data. Model diagnostic checking was carried out on ACF and PACF in residual terms through runs above and below the median, runs up and down and Ljung-Box tests. It is inferred that ARIMA (2,1,1) was found to be optimal and that the forecast values for the years 2019-20 to 2023-24 were estimated on the basis of this model, which were 7.82,8.23,8.72,9.2 and 9.72 million tonnes for the year 2019-20 to 2023-24, respectively. The significance of the mode is that we can forecast the values using this best fit model and forecast values are very important for the policymakers and other government agencies for proper policy decision regarding food security.

Meta-Learning Framework with Applications to Zero-Shot Time-Series Forecasting

Can meta-learning discover generic ways of processing time series (TS) from a diverse dataset so as to greatly improve generalization on new TS coming from different datasets? This work provides positive evidence to this using a broad meta-learning framework which we show subsumes many existing meta-learning algorithms. Our theoretical analysis suggests that residual connections act as a meta-learning adaptation mechanism, generating a subset of task-specific parameters based on a given TS input, thus gradually expanding the expressive power of the architecture on-the-fly. The same mechanism is shown via linearization analysis to have the interpretation of a sequential update of the final linear layer. Our empirical results on a wide range of data emphasize the importance of the identified meta-learning mechanisms for successful zero-shot univariate forecasting, suggesting that it is viable to train a neural network on a source TS dataset and deploy it on a different target TS dataset without retraining, resulting in performance that is at least as good as that of state-of-practice univariate forecasting models.

Prediction of Landsliding using Univariate Forecasting Models

In last few decades, many methods are proposed for time‐series forecasting. As always, when alternatives exists, choice needs to be made so that an appropriate forecasting method can be selected, and used for a specific forecasting. Primarily, the type of data used for time‐series forecasting are univariate and multivariate. In this paper, we presented an analysis of univariate time‐series forecasting data using ARIMA, GARCH and Dynamic Neural Network (DNN) modeling techniques. These techniques depend on a variety of parameters such as objective of forecasting, type of forecasted data and whether an automatic or manual approach is to be used for forecasting. We implemented proposed methods for 15 m landslide sensor data. The objective of the paper is to find a best method among well‐known techniques for landslide forecasting. The obtained results validate that by implementing three different models, DNN is best‐in‐class for time‐series landslide forecasting.

Performance of Electricity Price Forecasting Models: Evidence from Turkey

ABSTRACTIn this article, hourly prices of the Turkish Day Ahead Electricity Market are forecasted by using various univariate electricity price models, then the out-of-sample forecasts are compared with each other and the benchmarks. This article has two main contributions to the literature: Firstly, it provides a factorial Analysis of Variance (ANOVA) as a pre-whitening method of the price series and allows one to work with the stationary residuals series. Secondly, it is the first work, which compares the performances of all important statistical univariate forecast models in the Turkish electricity market. Results indicate the importance of the factorial ANOVA application and the SARIMA model’s success under the given conditions.

Forecasting the Incidence and Prevalence of Patients with End-Stage Renal Disease in Malaysia up to the Year 2040.

Background. The incidence of patients with end-stage renal disease (ESRD) requiring dialysis has been growing rapidly in Malaysia from 18 per million population (pmp) in 1993 to 231 pmp in 2013. Objective. To forecast the incidence and prevalence of ESRD patients who will require dialysis treatment in Malaysia until 2040. Methodology. Univariate forecasting models using the number of new and current dialysis patients, by the Malaysian Dialysis and Transplant Registry from 1993 to 2013 were used. Four forecasting models were evaluated, and the model with the smallest error was selected for the prediction. Result. ARIMA (0, 2, 1) modeling with the lowest error was selected to predict both the incidence (RMSE = 135.50, MAPE = 2.85, and MAE = 87.71) and the prevalence (RMSE = 158.79, MAPE = 1.29, and MAE = 117.21) of dialysis patients. The estimated incidences of new dialysis patients in 2020 and 2040 are 10,208 and 19,418 cases, respectively, while the estimated prevalence is 51,269 and 106,249 cases. Conclusion. The growth of ESRD patients on dialysis in Malaysia can be expected to continue at an alarming rate. Effective steps to address and curb further increase in new patients requiring dialysis are urgently needed, in order to mitigate the expected financial and health catastrophes associated with the projected increase of such patients.

Residential property prices’ modeling: evidence from selected European countries

PurposeThe purpose of this study is to contribute to the relatively narrow existing residential real estate literature by developing and validating several univariate forecasting models, to reliably anticipate future house price dynamics across several European Union (EU) countries.Design/methodology/approachThe research approach relies on the time series analysis, by using the Box–Jenkins autoregressive integrated moving average (ARIMA) methodology to explore the trends of residential property prices in selected EU countries and to obtain a snapshot of the potential signs of change to be witnessed by domestic residential markets on a short time-period. The analysis has been performed distinctly for each country in the sample, to account for country-specific past and future trends as well as similarities in their house price growth rate evolutions. The models were estimated for a broad sample of quarterly observations during 1990-2015, while the forecast horizon ranged between the third quarter of 2015 and the fourth quarter of 2016.FindingsThe findings suggested that residential property prices’ real growth rate can be modeled through the Box–Jenkins method for France, The Netherlands, Sweden and UK. The pattern of Italy’s residential property prices’ real growth rate cannot be explained by means of univariate ARIMA models, being more suited for multivariate models.Originality/valueThe article subscribes to the need for timely, high-frequency and quality data about house price trends in Europe, to increase the accuracy of forecasts and prevent the appearance of bubbles on real estate market. It compares residential property prices’ dynamics across European countries to identify housing markets with similar patterns of their prices.

Forecasting Inflation Using the Phillips Curve: Evidence from Swedish Data

This paper studies the forecasting ability of various Phillips curve specifications in a pseudo out-of-sample exercise for Swedish inflation over the period 1980–2014. Three measures of inflation are considered––headline inflation, underlying inflation, GDP deflator inflation, in addition to different activity variables, various econometric specifications and different sample periods. Although the results indicate heterogeneity in individual model performance and evidence of model instability, in general, the Phillips curve models improve inflation forecasts against the random walk benchmark for both headline inflation and underlying inflation, and fail to beat the random walk benchmark for GDP deflator inflation. Phillips curve forecasts beat the random walk benchmark especially for 2004–2013. The monetary regime change in 1993 from exchange rate targeting to inflation targeting is also taken into account. The results suggest that for all Phillips curve models and all three inflation measures, the performance of the Phillips curve depends on whether the data used for making the predictions was under the inflation targeting regime or not. Univariate forecasting models perform well for the fixed exchange rate regime period, but the Phillips curve models are useful under the inflation targeting regime.

Lasso-Based Forecast Combinations for Forecasting Realized Variances

Volatility forecasts are key inputs in financial analysis. While lasso based forecasts have shown to perform well in many applications, their use to obtain volatility forecasts has not yet received much attention in the literature. Lasso estimators produce parsimonious forecast models. Our forecast combination approach hedges against the risk of selecting a wrong degree of model parsimony. Apart from the standard lasso, we consider several lasso extensions that account for the dynamic nature of the forecast model. We apply forecast combined lasso estimators in a comprehensive forecasting exercise using realized variance time series of ten major international stock market indices. We find the lasso extended 'ordered lasso' to give the most accurate realized variance forecasts. Multivariate forecast models, accounting for volatility spillovers between different stock markets, outperform univariate forecast models for longer forecast horizons.

Forecasting construction output: a comparison of artificial neural network and Box-Jenkins model

Purpose – Fluctuations in construction output has an adverse effect on the construction industry and the economy due to its strong linkage. Developing reliable and accurate predictive models is vital to implementing effective response strategies to mitigate the impact of such fluctuations. The purpose of this paper is to compare the accuracy of two univariate forecast models, i.e. Box-Jenkins (autoregressive integrated moving average (ARIMA)) and Neural Network Autoregressive (NNAR). Design/methodology/approach – Four quarterly time-series data on the construction output of Hong Kong were collected (1983Q1-2014Q4). The collected data were divided into two parts. The first part was fitted to the model, while the other was used to evaluate the predictive accuracy of the developed models. Findings – The NNAR model can provide reliable and accurate forecast of total, private and “others” construction output for the medium term. In addition, the NNAR model outperforms the ARIMA model, in terms of accuracy. Research limitations/implications – The applicability of the NNAR model to the construction industry of other countries could be further explored. The main limitation of artificial intelligence models is the lack of explanatory capability. Practical implications – The NNAR model could be used as a tool for accurately predicting future patterns in construction output. This is vital for the sustained growth of the construction industry and the economy. Originality/value – This is the first study to apply the NNAR model to construction output forecasting research.

Do Phillips Curves Conditionally Help to Forecast Inflation?

This paper reexamines the forecasting ability of Phillips curves from both an unconditional and conditional perspective by applying the method developed by Giacomini and White (2006). We find that forecasts from our Phillips curve models tend to be unconditionally inferior to those from our univariate forecasting models. We also find, however, that conditioning on the state of the economy sometimes does improve the performance of the Phillips curve model in a statistically significant manner. When we do find improvement, it is asymmetric -- Phillips curve forecasts tend to be more accurate when the economy is weak and less accurate when the economy is strong. Any improvement we find, however, vanished over the post-1984 period.

The importance of climatic factors and outliers in predicting regional monthly campylobacteriosis risk in Georgia, USA

Incidence of Campylobacter infection exhibits a strong seasonal component and regional variations in temperate climate zones. Forecasting the risk of infection regionally may provide clues to identify sources of transmission affected by temperature and precipitation. The objectives of this study were to (1) assess temporal patterns and differences in campylobacteriosis risk among nine climatic divisions of Georgia, USA, (2) compare univariate forecasting models that analyze campylobacteriosis risk over time with those that incorporate temperature and/or precipitation, and (3) investigate alternatives to supposedly random walk series and non-random occurrences that could be outliers. Temporal patterns of campylobacteriosis risk in Georgia were visually and statistically assessed. Univariate and multivariable forecasting models were used to predict the risk of campylobacteriosis and the coefficient of determination (R 2) was used for evaluating training (1999–2007) and holdout (2008) samples. Statistical control charting and rolling holdout periods were investigated to better understand the effect of outliers and improve forecasts. State and division level campylobacteriosis risk exhibited seasonal patterns with peaks occurring between June and August, and there were significant associations between campylobacteriosis risk, precipitation, and temperature. State and combined division forecasts were better than divisions alone, and models that included climate variables were comparable to univariate models. While rolling holdout techniques did not improve predictive ability, control charting identified high-risk time periods that require further investigation. These findings are important in (1) determining how climatic factors affect environmental sources and reservoirs of Campylobacter spp. and (2) identifying regional spikes in the risk of human Campylobacter infection and their underlying causes.

Forecasting Demand for Australian Passports

We forecast demand for Australian passports using a number of univariate and multivariate forecasting models, and assess their relative predictive ability over a number of forecasting horizons and evaluation measures. Our key result is to use different forecasting models for predicting passport demand in the short- versus medium- to long-run. Specifically, to forecast Adult-and-Senior passport demand in the short-term (i.e. up to 12 months) univariate ARIMA models are preferred, while for the longer term forecasts multivariate models with exogenous variables outperform, although only marginally. To forecast passport demand for Minors (less than 18 years old) ARIMA models perform well both in the short-term and the long-term, although ARIMA with explanatory variables outperforms slightly.

Comment

Comment

Do Phillips Curves Conditionally Help to Forecast Inflation?

This paper reexamines the forecasting ability of Phillips curves from both an unconditional and conditional perspective by applying the method developed by Giacomini and White (2006). We find that forecasts from our Phillips curve models tend to be unconditionally inferior to those from our univariate forecasting models. We also find, however, that conditioning on the state of the economy sometimes does improve the performance of the Phillips curve model in a statistically significant manner. When we do find improvement, it is asymmetric -- Phillips curve forecasts tend to be more accurate when the economy is weak and less accurate when the economy is strong. Any improvement we find, however, vanished over the post-1984 period. Supersedes WP 11-40.