Multivariate Time Series Regression Research Articles

Impact of grid size on spatiotemporal prediction of fine particulate matter

Incorporating high spatiotemporal resolution in prediction models can significantly improve the accuracy of results. The current study focuses on identifying the most suitable spatial resolution using hexagonal tessellations of different sizes over the study area. Hexagonal grids are selected after comparing different types of grids. Spatial interpolation methods, including inverse distance weighting and ordinary Kriging method, are used for creating augmented datasets for each grid. A hybrid deep-learning-based prediction model is developed combining Convolutional Neural Network (CNN) and Long Short-term Memory (LSTM) for multivariate time series regression to predict PM2.5. Three scenarios are considered based on the size of the hexagonal grids. The results from each model are compared using evaluation indices of Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The R2 values range from 0.8 to 0.95, which shows that the hybrid model gives accurate predictions. Analysis of models for different-sized grids shows that for 1, 3, and 5 km grid sizes, 17%, 24%, and 6% of the cells have RMSE values less than 20, respectively. Similarly, 39%, 50%, and 46% of the cells have MAE values less than 16 for 1, 3, and 5 km grid sizes, respectively. The comparison showed that the 3 km grid benefits data aggregation per prediction accuracy and computational time. The predictive capability of models is not dependent on the spatial relationship between data from different static monitors. In the future, mobile monitoring data can be included to increase the data points and models’ accuracy.

Impacts of Atmospheric and Load Conditions on the Power Substation Equipment Temperature Model

Infrared thermography is a predictive maintenance tool used in substations to identify a disturbance in electrical equipment that could lead to poor operation and potential failure in the future. According to Joule’s law, the temperature of electrical equipment is proportional to the current flowing through it. Other external factors, such as solar incidence, air humidity, wind speed, and air temperature, can interfere with its operating temperatures. Based on this premise, this article aims to analyze the influence of atmospheric and load conditions on the operational cycle of thermography-monitored equipment in order to describe the operating temperature of the object using only external data and to show the impacts of external influences on the final temperature reached by the object. Five multivariate time series regression models were developed to describe the maximum equipment temperature. The final model achieved the best fit between the measured and model temperature based on the Akaike information criterion (AIC) metric, where all external variables were used to compose the model. The proposed model shows the impacts of each external factor on equipment temperature and could be used to create a predictive maintenance strategy for power substations to avoid failure.

Graph neural networks for multivariate time series regression with application to seismic data

Machine learning, with its advances in deep learning has shown great potential in analyzing time series. In many scenarios, however, additional information that can potentially improve the predictions is available. This is crucial for data that arise from e. g., sensor networks that contain information about sensor locations. Then, such spatial information can be exploited by modeling it via graph structures, along with the sequential (time series) information. Recent advances in adapting deep learning to graphs have shown potential in various tasks. However, these methods have not been adapted for time series tasks to a great extent. Most attempts have essentially consolidated around time series forecasting with small sequence lengths. Generally, these architectures are not well suited for regression or classification tasks where the value to be predicted is not strictly depending on the most recent values, but rather on the whole length of the time series. We propose TISER-GCN, a novel graph neural network architecture for processing, in particular, these long time series in a multivariate regression task. Our proposed model is tested on two seismic datasets containing earthquake waveforms, where the goal is to predict maximum intensity measurements of ground shaking at each seismic station. Our findings demonstrate promising results of our approach—with an average MSE reduction of 16.3%—compared to the best performing baselines. In addition, our approach matches the baseline scores by needing only half the input size. The results are discussed in depth with an additional ablation study.

Dataset for: Multivariate Time Series Regression with Graph Neural Networks

Dataset for: Multivariate Time Series Regression with Graph Neural Networks

Focusing hospitalist roles on either admitting or rounding facilitates unit-based assignment and is associated with improved discharge efficiency

ABSTRACT Background Unit-based teams may improve care delivery for hospitalized patients but can be challenging to implement broadly across all acute care units in a hospital. Objective To determine the effect of a Lean-guided transition to hospital-wide unit-based assignment on care delivery outcomes. Design, setting, and participants The study was a retrospective time-series with primary outcomes of discharge efficiency, 30-day readmissions, and length of stay, performed at a 336-bed tertiary academic referral hospital in the Pacific Northwest with approximately 17,000 admissions annually. Intervention Implementation of a Lean-guided quality improvement intervention included division of hospitalist duties into ‘admitters’ and ‘rounders,’ with simulated patient flow exercises to determine the optimal staffing model. Main outcomes and measures Discharge efficiency (number of patients discharged by hospitalists divided by the number of hospitalist patient encounter days per month) and 30-day readmissions were compared using the t-test or chi-square, and length of stay was analyzed in a multivariate time-series regression model. Results The intervention was associated with a significant improvement in discharge efficiency, by 0.014 (from 0.168 to 0.181) discharges/encounter (95% CI = 0.024, 0.004), p = 0.009. Mean length of stay decreased by 0.98 days (95% CI 0.50, 1.47) after adjustment for patient age, patient type (medical versus surgical), critical care admissions, and discharge disposition, without a corresponding change in 30-day readmission rate (12.2% (1948/15,902) pre-intervention to 11.7% (397/3379) post-intervention (p = 0.42)). Conclusions Dividing hospitalist roles into admitters and rounders enabled implementation of unit-based teams across the hospital, with corresponding improvements in discharge efficiency and length of stay.

EFFECT OF EXTERNAL DEBT LIABILITY ON ECONOMIC GROWTH IN KENYA

Purpose: The purpose of the study was to evaluate the effect of external debt liability on economic growth in Kenya.
 Materials and Methods: The descriptive research design was adopted. The target population was three institutions: The National Treasury, Kenya National Bureau of Statistics, and the World Bank. The study used time series data. The designated sample for this study covered a period of 43 years (1977–2019). Secondary data was used in this study. The data collected was on GDP of Kenya between 1977 and 2019, External public debt in terms of US dollars from 1977 to 2019, External private debt from 1977 and 2019 and external debt service payments from 1977 to 2019, all in US dollars. A data collection sheet was used to collect the data on the four variables. World Bank and World Development Indicator economic Meta data and published data by Central Bank of Kenya and the Kenya National Bureau of Statistics were the source of data for this study. The study used Eviews version 10 for analyzing and presenting study findings. The study employed multivariate time series and panel data regression analysis. The model employed GDP as a measure of economic growth and external public debt, external private debt, and external debt service payment as its main independent variables.
 Results: The study found out that only the external private debt and the debt service payment showed bilateral causal relationship. External public debt and external private debt had a positive and significant effect on the GDP, indicating that external debt promotes economic growth in Kenya. The external debt service payment showed a negative and a significant effect on the GDP as well. The model explained 97% variability of the GDP as explained by the three independent variables combined. The 3% is attributed to other factors, not included in this study.
 Unique contribution to theory, practice and policy: The study recommends a more robust multivariate model to be employed to include more macro-economic variables to explain economic growth. A decade-to-decade comparison can also be done to compare the effects of the external debt on Kenyan economic growth in different time intervals. Fiscal and monetary policies should be reviewed to encourage more domestic and foreign investments and discourage external borrowing to fund budget deficits or projects with low or no returns.

The Effect of Climate Variables on the Incidence of Cutaneous Leishmaniasis in Isfahan, Central Iran

BACKGROUND AND AIM: In recent years, there have been considerable changes in the distribution of diseases that are potentially tied to ongoing climate variability. The aim of this study was to investigate the association between incidence of cutaneous leishmaniasis (CL) and climatic factors in an Iranian city (Isfahan) which had the highest incidence of CL in the country. METHODS: CL incidence and climate data were inquired from April 2010 to March 2017 (108 months) for Isfahan city. Univariate and multivariate Seasonal Autoregressive Integrated Moving Average (SARIMA), Generalized Additive Models (GAM) and Generalized Additive Mixed Models (GAMM) were used to identify the association between CL cases and climate variables, and forecast CL incidence. AIC, BIC and residual tests were used to test the goodness of fit of SARIMA models; and R2 was used for GAM/GAMM. RESULTS:5297 CL cases were recorded during this time. The incidence had a seasonal pattern and the highest number of cases were recorded from August to November. In univariate SARIMA, (1,0,1) (0,1,1)12 was the best fit for predicting CL incidence (AIC=8.09, BIC=8.32). Multivariate time series regression (1,0,1)(0,1,1)12 showed that monthly mean humidity after 3 months lag was inversely related to CL incidence (β=-1.59, p=0.0072, AIC=8.52, BIC=8.66). GAMM results showed average temperature with 2-month lag, average relative humidity with 4-month lag, monthly cumulative rainfall with 1-month lag and monthly sunshine hours with 1-month lag were related to CL incidence (R2=0.94). CONCLUSIONS:The impact of climate variables on the incidence of Leishmania is not linear and GAM models that include non-linear structures are a better fit for prediction. In Isfahan, Iran, climate variables can greatly predict the incidence of CL and these variables can be used for predicting outbreaks. KEYWORDS: Time-series analysis; SARIMA; Leishmaniasis, Iran

A novel wide & deep transfer learning stacked GRU framework for network intrusion detection

With the increasing frequency, severity and complexity of recent cyber attacks around the world, network intrusion detection has become mandatory and highly sophisticated task. Achieving high performance in network intrusion detection by applying benchmark machine learning classifiers (including deep learning techniques) has become a major challenge in recent times. One of the biggest challenges is improving the memorization capacity and generalization ability of NIDS (Network Intrusion Detection Systems). In this paper, we propose a highly scalable novel wide & deep transfer learning (TL) based stacked GRU (Gated Recurrent Unit) model to deal with multi-dimensional point data and multi-variate time series regression and classification problems in network intrusion detection. The proposed model has the memorization capacity of linear regression model and the generalization ability of deep GRU model. The deep component consists of a transfer learning framework that pretrains a source model and then fine-tunes the whole source model on the same dataset multiple times until it gives peak performance. This method gives a multi-class evaluation accuracy of 99.92% on KDDCup 99(10%) dataset and 94.22% on UNSW-NB15 dataset respectively. Extensive experimentations and evaluations have been carried out by comparing it with other machine learning (including deep learning) network intrusion detection techniques. The proposed method outperforms most of the existing intrusion detection approaches.

МАТЕМАТИЧНІ МЕТОДИ МОДЕЛЮВАННЯ ТА ПРОГНОЗУВАННЯ РИНКУ ПРАЦІ

The labor market is a dynamic system that requires constant monitoring and analysis. To determine the vector of activity of subjects of the labor market and information about market conditions, the need for modeling and forecasting key indicators of the labor market increases. No less important is the methodological collection, analysis and comparison of information related to the functioning of the labor market, forecasting and planning of market activities for further decision-making in management. In modern publications on economics, devoted to the analysis of the labor market, not enough attention is paid to studies that generally cover all methods of economic and mathematical modeling and management of the labor market. Most often, there are scientific works that are related only to control in a specific plane, where one or several methods are applied simultaneously. The publication highlights the main areas of labor market research and methods of economic and mathematical modeling, which are advisable to use in describing the state and dynamics of the sphere of labor relations. The main principles of economic and mathematical modeling in assessing the development of the labor market are highlighted: the consistency of forecasting, the correspondence of forecasting to modern scientific research, the adequacy of forecasts to real processes, the scenario nature of forecasts. The characteristics of economic, mathematical and statistical methods and models that can be used to manage the labor market are given. The paper presents a brief description and the possibility of applying in labor market research such methods as: the system analysis method, aggregate indexing, the indicative method, the comparison method, the assessment of external and internal factors, statistical methods based on the calculation of the standard deviation, variance, correlation coefficients and variations, statistical multivariate analysis, balance models, trend and time series analysis, correlation and regression analysis, simulation modeling, neural networks, sociological research, the method of expert estimates and others. The main software packages for the automation of the above methods are also systematized.

Economic Forces and the Stock Market Performance in Developing Countries: Evidence From Sudan

This paper is an attempt to empirically investigate the determinants of the stock market performance in Sudan. It aims at identifying the short and long run relationships between the Khartoum Stock Exchange all- share price index (KSI) as an indicator of market performance and some selected micro and macro-economic factors. The inflation rate, cost of capital, foreign exchange rate, broad money supply, and crude oil price are chosen as proxies for macroeconomic factors. The market oriented indicators used include market capitalization, market trading system, and market trading volume. The study covers the period 2003-2017. The paper employs the Multivariate Time Series Regression Analysis to estimate the short run relationship between the selected independent variables and the KSE price index. Soren Johansen’s Cointegration Test and Vector Error Correction Model (VECM) have been employed to identify the long run equilibrium relationship among the variables. To estimate the causal relationship between the selected variables Toda-Yamamoto (T-Y) Granger Causality Test has been utilized. The study documents that the Khartoum Stock Exchange performance is significantly affected both by micro and macroeconomic factors. In the long run, all the independent variables with the exception of the cost of capital, have a significant positive relationship with KSI. However, in the short run the determinants of the stock market performance are market capitalization, market trading volume, money Supply, and cost of capital.

The role of GRACE total water storage anomalies, streamflow and rainfall in stream salinity trends across Australia’s Murray-Darling Basin during and post the Millennium Drought

By influencing water tables of saline aquifers, multiyear dry or wet periods can significantly delay or accelerate dryland salinity, but this effect remains poorly quantified at the large river basin scale. The Gravity and Climate Recovery Experiment (GRACE) satellite measures changes in the total water storage of river systems, providing a unique opportunity for better understanding connections between stream salinity and changes in catchment water storages at the large river basin scale. Here, we quantified the role of GRACE total water storage anomalies (TWSA) in stream salinity variability across Australia’s Murray-Darling Basin (∼1 million km2), while also accounting for streamflow and rainfall. We used the MERRA-2 global land surface model to i) place our findings in the context of the longer-term hydroclimatology (1980-present) and ii) to decompose TWSA into groundwater storage as an alternative driver variable. Multivariate time series regression models (generalized additive mixed models or GAMM) showed that the driver variables could explain 20–50% of the variability in stream salinity across 8 sub-catchments in the Murray Darling Basin. TWSA commonly explained as much variability as streamflow, while groundwater storage and TWSA had very similar explanatory power and rainfall only negligible contributions. The 2000–2009 Millennium Drought and the subsequent La Nina Floods had a predominantly decelerating and accelerating effect on stream salinity respectively and these trends were partially explained by trends in TWSA. Our study illustrates that GRACE can be a useful addition for monitoring and modeling dryland salinity over large river basins.

A Bayesian approach to developing a strategic early warning system for the French milk market

A new approach is provided in our paper for creating a strategic early warningsystem allowing the estimation of the future state of the milk market as scenarios. This is inline with the recent call from the EU commission for tools that help to better address such ahighly volatile market. We applied different multivariate time series regression and Bayesiannetworks on a pre-determined map of relations between macro-economic indicators. Theevaluation of our findings with root mean square error (RMSE) performance score enhancesthe robustness of the prediction model constructed. Our model could be used by competitiveintelligence teams to obtain sharper scenarios, leading companies and public organisations tobetter anticipate market changes and make more robust decisions.

Internal and External Factors of Future Returns in the Banking Business: Time Series Analysis of Interrelationship

The paper seeks to find the interrelationship between internal and external factors of future returns in the banking business. A multivariate time series regression models are fitted for the dependent variable: return on equity (ROE) against the lag one independent variables, namely: deposit, size, loan, capital, inflation, gross domestic product (GDP) and stock market capitalization (SMC), for ABSA bank; using secondary data, which span from 1998 to 2014 fiscal years. Logarithm transformation of the absolute value of the de-trended data and first differencing at lag one were the smoothing techniques applied to the data. Multivariate time series regression by the least square approach with special consideration of the stepwise method was used in fitting the models to the data. Results indicated that first, there is a positive linear relationship between ROE and loans, a negative linear relationship between ROE and inflation from the differencing techniques; and equally a negative log-linear relationship between ROE and capital as well as a positive log-linear relationship between ROE and ROA for the logarithm de-trend technique.

Internal and External Factors of Future Returns in the Banking Business: Time Series Analysis of Interrelationship

The paper seeks to find the interrelationship between internal and external factors of future returns in the banking business. A multivariate time series regression models are fitted for the dependent variable: return on equity (ROE) against the lag one independent variables, namely: deposit, size, loan, capital, inflation, gross domestic product (GDP) and stock market capitalization (SMC), for ABSA bank; using secondary data, which span from 1998 to 2014 fiscal years. Logarithm transformation of the absolute value of the de-trended data and first differencing at lag one were the smoothing techniques applied to the data. Multivariate time series regression by the least square approach with special consideration of the stepwise method was used in fitting the models to the data. Results indicated that first, there is a positive linear relationship between ROE and loans, a negative linear relationship between ROE and inflation from the differencing techniques; and equally a negative log-linear relationship between ROE and capital as well as a positive log-linear relationship between ROE and ROA for the logarithm de-trend technique.

Stage-specific effects of the Sarbanes-Oxley Act

Purpose – The purpose of this paper is to empirically demonstrate that drivers of venture capital (VC) investments are different across three broadly defined sectors: high-technology manufacturing, medium-technology manufacturing and services, and low-technology services. Moreover, such differences also exist across industries within each of these sectors. Design/methodology/approach – The basic hypothesis is that “not only different stages of VC investments have different drivers, but VC investments in different sectors of the economy are also driven by different drivers.” The paper tests this hypothesis using a Poterba (1989) type supply and demand framework in the multivariate time-series regression analysis. Findings – This paper empirically demonstrates that drivers of VC investments are different across three broadly defined sectors: high-technology manufacturing, medium-technology manufacturing and services, and low-technology services. Moreover, such differences also exist by stages of investment and across industries within each of these sectors. In particular, the paper finds that the importance of the number of VC-led initial public offering (IPO) transactions as the main driver of VC investment decreases with the level of technology involved in the sector. IPO transactions are particularly important in software, networking and equipment, and business products and services industries. In contrast to earlier literature, however, the paper do not find a more pronounced effect of IPOs for seed and late stages of VC investments. Similarly, the positive impact Sarbanes-Oxley Act of 2002 – which mainly impacts public companies – also intensifies with a decrease in the level of technology involved in the sector, and the paper do not find a negative impact. The Act is important particularly for VC investments in medium- and low-tech sectors and in early or expansion stages. Originality/value – In analyzing the determinants of VC in a supply and demand framework as in Poterba (1989), the paper differentiates between different sectors (17 industries) and stages of VC (four stages: seed, early, expansion, late). Such level of differentiation is novel and allows more refined and better targeted public policy measures.

Fever in Adult ICUs

Fever is common and associated with increased mortality among patients admitted to adult ICUs, yet recent literature suggests that the incidence of fever may be decreasing. The objective of this study was to determine whether the incidence of fever in adult ICUs changed over time and the factors responsible for the observed change. Interrupted time series analysis. The primary outcome was the cumulative incidence of fever (temperature ≥ 38.3 °C). Secondary outcomes included the cumulative rate of blood cultures ordered, and the cumulative incidence of bloodstream infections and ventilator-associated pneumonia. Data were analyzed with segmented linear regression and adjusted for important confounding variables. Calgary zone of Alberta Health Services between January 1, 2004, and June 30, 2009. Adults (age ≥ 18 yr) admitted to ICUs. None. There were 18,989 ICU admissions among 17,153 patients. The cumulative incidence of fever during ICU admission decreased from 50.1% of all patients to 25.5% over the 5.5-year study period. Implementation of a new noninvasive thermometer was associated with a 5.1% (95% CI, 1.4-8.9%, p = 0.01) absolute decrease in fever incidence; however, the decrease in fever incidence was predominantly a function of a constant baseline decrease of 1.1% per quarter (95% CI, 0.8-1.5%, p < 0.0001). Multivariate logistic time series regression found that time and thermometer change were the only independent predictors of the changing incidence of fever. The ordering of blood cultures, bloodstream infection incidence, and ICU mortality were unchanged throughout the study period. The incidence of fever in adult ICUs decreased considerably over time. The lack of change in the ordering of blood cultures and the incidence of bloodstream infections calls into question the importance of fever during the diagnostic evaluation of critically ill patients.

Investor Sentiment and Stock Return: Do Industries Matter?

The basic objective of this article is to evaluate the pricing implications of market-wide investor sentiment risk for cross-sectional return variations of Indian listed companies across industry groups. A multivariate time-series regression approach has been used to examine the impact of sentiment risk on stock return behaviour in the presence of other market-wide systematic risk factors. Our results suggest that the role of sentiment risk in the determination of a cross-section of stock returns is not uniform across the test asset portfolios formed on the basis of size, book-to-market equity, liquidity and momentum characteristics. For all portfolios, the impact of sentiment risk on the cross-section of stock returns behaviour has been disproportionately negative. The effect holds even after controlling for systematic market-wide risk factors. Although the impact of sentiment risk on industry-shorted portfolio returns persists in accordance with the theoretical argument, the cross-sectional variation with respect to different industries has been heavily dependent on the availability of stocks in that particular industry. The commonality of the sentiment effect across industry is not similar, as it is for the aggregate market. The results suggest that generalisation of the hard-to-value and difficult-to-arbitrage argument must be judged with caution, keeping the industry effects in mind. JEL Classification: GI,G12,G14

Rendición de cuentas horizontal y preponderancia presidencial: una combinación difícil. El caso de República Dominicana, 1967-2009

Este artículo estudia la rendición de cuentas horizontal en República Dominicana. Primero se estudian las normas constitucionales que regulan el sistema de pesos y contrapesos. Aunque importantes, estas no explican la gran variación en niveles de rendición de cuentas observada en el caso dominicano. Por lo tanto, el trabajo se enfoca en el ejercicio de los gastos de los presupuestos nacionales y la situación financiera de las instituciones de control. El artículo sugiere analizar la situación presupuestaria de las instituciones fiscalizadoras y propone utilizar la desproporcionalidad entre el presupuesto aprobado y el ejecutado como un indicador del nivel de rendición de cuentas en un país. Un análisis multivariado de regresión de series de tiempo demuestra que la desproporcionalidad presupuestal aumenta con la preponderancia presidencial. Se concluye que, para el caso dominicano en el período 1967-2009, el comportamiento de varios presidentes más que otros factores es el que ha obstaculizado el funcionamiento del sistema de pesos y contrapesos. Por lo tanto, la experiencia dominicana confirma la hipótesis de O’Donnell sobre las democracias delegativas.

Exploring Reductions in London Underground Passenger Journeys Following the July 2005 Bombings

We examine the reduction in London Underground passenger journeys in response to the July 2005 bombings. Using entrance data for London Underground stations between 2001 and 2007, we incorporate demand and supply factors in a multivariate time-series regression model to estimate changes in passenger journeys between different Underground lines. We find that passenger journeys fell by an average of 8.3% for the 4 months following the attacks. This amounts to an overall reduction of 22.5 million passenger journeys for that period. Passenger journeys returned to predicted levels during September 2005, yet we find evidence of reduced travel until June 2006. Our estimates controlled for other factors, including reduced Underground service provision due to damage from the attacks, economic conditions, and weather, yet substantial reductions in passenger journeys remained. Around 82% of passenger journey reductions following the 2005 attacks cannot be attributed to supply-side factors or demand-side factors such as economic conditions, weather, or the summer school-break alone. We suggest that this reduction may partially be due to an increased perception of the risk of Underground travel after the attacks.

Assessment of varicella vaccine effectiveness in Germany: a time-series approach

A multivariate time-series regression model was developed in order to describe the 2005-2008 age-specific time-course of varicella sentinel surveillance data following the introduction of a varicella childhood vaccination programme in Germany. This ecological approach allows the assessment of vaccine effectiveness under field conditions by relating vaccine coverage in cohorts of 24-month-old children to the mean number of cases per reporting unit in the sentinel network. For the 1-2 years age group, which is directly affected by the vaccination programme, a one-dose vaccine effectiveness of 83·2% (95% CI 80·2-85·7) was estimated which corresponds to previous approaches assessing varicella vaccine effectiveness in the field in the USA.