



































Type of the Paper (Article


American Interdisciplinary Journal of Business 

and Economics 
ISSN: 2837-1909| Impact Factor : 4.6 

Volume. 9, Number 4; October-December, 2022; 

Published By: Scientific and Academic Development Institute (SADI) 

8933 Willis Ave Los Angeles, California 

https://sadipub.com/Journals/index.php/aijbe 

 

 

8 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

PROPHET MODEL'S EFFICIENCY IN SHORT-TERM COVID-19 

CUMULATIVE CASE PROJECTIONS: G7 COUNTRIES 

 

 

Yeliz Zoğal 
Ankara Haci Bayram Veli University 

 

Abstract: The Covid-19 pandemic has had a significant impact on the health and well-being of people across 
the globe, as well as the global economy at large. It has become essential to predict the spread of infectious 
diseases like Covid-19 to understand its impact on public health and the economy. This study analyzes short-
term predictions of Covid-19 cases in G7 countries using the Prophet model. The model uses a trend function, 
a seasonality function, and a holiday function to generate accurate short-term predictions. The study compares 
the predictions for G7 countries and finds that Canada and Germany had the lowest root mean square error 
(RMSE) values. The economic and financial impacts of the pandemic on global supply chains, job losses, and 
business closures are also analyzed. The study highlights the significant increase in public debt due to large-
scale fiscal stimulus packages implemented by governments to mitigate the economic impact of the pandemic. 
The study emphasizes that accurate predictions of the spread mechanism are crucial for managing the 
pandemic effectively and mitigating its impact. The literature review of various models indicates the 
importance of accurate predictions and the difficulties in creating them. The study recommends the use of 
machine learning models like the Prophet model to generate accurate short-term predictions to combat the 
Covid-19 pandemic's spread. 

Keywords: Covid-19, Prophet model, short-term predictions, G7 countries, root mean square error, economic 
impacts, global supply chains, fiscal stimulus packages, machine learning models.  

 

 1. Introduction  

In December 2019, it was reported to the world that a virus associated with severe acute respiratory syndrome 
began to spread across China's Wuhan city. This virus, later named Covid-19, has started to spread worldwide, 
and cases have become unpreventable. In the WHO Director General’s statement on COVID-19 dated March 
11, 2020, he stated that in two weeks, the number of Covid-19 cases increased thirteen times, the number of 
affected countries tripled, there were more than 118,000 cases in 114 countries and 4,291 people died.  
The unpreparedness of countries for infectious diseases has caused them to struggle with a lack of capacity, 
resources, and determination in general, especially in the health sector. In addition to the severe impact of 
COVID-19 on healthcare systems, the pandemic has had a significant and rapidly escalating impact on the 
world economy and businesses.  
The COVID-19 pandemic has been ongoing for more than three years and continues to cause significant health 
and economic losses. According to official estimates, more than 6 million people have died from the virus, 
with studies estimating the actual death toll to be much higher, ranging from 16 to 20 million, which is 
approximately equal to that of World War I. According to the IMF’s World Economic Outlook (2022), the 
cumulative output loss from the pandemic through 2024 is projected to be about $13.8 trillion and it is likely 
that the actual loss will be even higher. According to the World Economic Outlook (WEO) report published 
in October 2021, the economic contraction for G7 countries is as follows: Germany at 4.6%, France at 8%, 
Italy at 8.9%, United Kingdom at 9.8%, United States at 4.3%, Canada at 5.3%, and Japan at 4.6%. In G7 



Yeliz Zoğal (2022) 
 

9 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

countries, the pandemic has led to a significant increase in unemployment and a decrease in economic activity. 
G7 countries have also seen a decline in consumer spending, particularly in sectors such as travel and tourism.  
The economic impact of the COVID-19 pandemic has been very severe and has occurred much more rapidly 
than the 2008 global financial crisis (GFC) and the Great Depression. The rapid
spread of the virus and the measures put in place to control it have led to widespread job losses and business 
closures, disruptions in global supply chains, and a decline in economic activity. In the 2008 GFC and the 
Great Depression, stock markets collapsed by 50% or more, credit markets froze up, massive bankruptcies 
followed, unemployment rates soared above 10%, and GDP contracted at an annualized rate of 10% or more, 
but all of this took around three years to play out. While in the current crisis, similarly dire macroeconomic 
and financial outcomes have materialized in a much shorter period, in some cases, just three weeks. The speed 
of the economic downturn caused by the COVID-19 pandemic is largely due to the rapid spread of the virus 
and the measures put in place to control it. The pandemic has also led to a significant increase in public debt 
as governments have implemented large-scale fiscal stimulus packages to mitigate the economic impact of the 
pandemic.  
The COVID-19 crisis has brought an unprecedented shock to the labor market and has led to an unemployment 
crisis. Restrictions and lockdowns implemented since March 2020, as well as the decline in demand caused 
by the pandemic, have led to millions of job losses around the world. Although some measures taken to prevent 
the spread of the virus have allowed some people to work from home, in most countries it has caused 
unemployment. The visible contraction in the economy is reflected in the unemployment rates. According to 
the IMF's report in 2021, the unemployment rate in G7 countries is Germany at 3.8%, France at 8%, Italy at 
9.3%, United Kingdom at 4.5%, United States at 8.1%, Canada at 9.6% and Japan at 2.8%. In addition to the 
hike in unemployment rates, the profound effect has started to increase income inequalities and poverty, which 
was estimated as more 71 million people as of March 2021.  
The outbreak of COVID-19 has had a significant impact on global supply chains. The disruption of 
transportation and production caused by lockdowns and other measures taken to contain the spread of the virus 
has led to shortages of goods and materials, delays in delivery times, and increased costs for businesses. 
Consumers have also been affected by the disruption of supply chains, with many facing shortages of essential 
goods and higher prices for products. The domino effect of broken supply chains is that it can cause a ripple 
effect throughout the economy, affecting not just producers and consumers but also other businesses that rely 
on them. This can lead to job losses, reduced economic activity, and other negative consequences.  
On the supply side, production chains have been disrupted, while on the demand side, consumption and 
investment spending have been negatively affected. This is likely to exacerbate the ongoing economic 
downturn, making it more pronounced. To address both the pandemic and economic downturn, governments 
have been urged to "Go big. Act fast. Keep the lights on" by economist Richard Baldwin, who argues that 
combining restrictive policies that reduce production with stimulus policies that maintain spending will create 
supply-side problems and lead to cost-driven inflation [2]. In other words, the idea that the global downturn 
can be revived only by increasing credit and borrowing more heavily for consumption is an illusion.  
The COVID-19 pandemic has had a significant impact on healthcare economics around the world. In the short 
run, healthcare facilities have been overwhelmed by the influx of patients, leading to increased costs for 
inpatient and outpatient care. This has been compounded by the need for additional resources such as personal 
protective equipment and additional staff to handle the increased workload. In the long run, the economic 
impact of the pandemic on healthcare systems may be even more severe. The prolonged disruption of 
healthcare services and the increased demand for care could lead to higher costs for both patients and 
healthcare providers, as well as longer wait times for appointments and procedures. Additionally, the pandemic 
has led to a decline in revenue for many healthcare providers and hospitals, which could lead to financial 
difficulties and closures. This could lead to further strain on the healthcare system in the long run as the 
population increases and aging.  
The listed reasons above make understanding the spread mechanism and forecasting essential for effectively 
managing the pandemic and minimizing its impact on public health and the economy.  
In the literature, many studies deal with understanding the spread mechanism. For instance, the mathematical 
SIR (Susceptible, Infected and Recovered) model provides differential solutions by dividing the total 



Yeliz Zoğal (2022) 
 

10 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

population into different groups. The model describes the flow of individuals between these compartments 
based on certain assumptions, such as the rate at which infected individuals infect susceptible individuals (the 
infection rate) and the rate at which infected individuals recover or are removed from the population (the 
removal rate). The SIR model can be used to estimate the number of individuals who will be infected and 
recovered over time, as well as the peak of the epidemic. Although the characteristics of the infectious disease 
shape the model, it does not provide satisfactory results in the early stages of the disease. The econometric 
time series model, ARIMA (autoregressive integrated moving average), provides more realistic results as the 
data increase since it predicts the future of the variable with past values. The ARIMA model can be used to 
model and forecast time series data with a trend and/or seasonality. The ARIMA model is widely used in 
various fields such as finance, economics, engineering, and infectious diseases, and it is considered as one of 
the most powerful tool for time series forecasting. In addition, machine learning models (supervised learning, 
reinforcement learning models, deep learning models, ensemble models) are frequently used to predict 
infectious diseases. The present study aims to provide a five-day Covid-19 prediction for G7 (Canada, France, 
Germany, Italy, Japan, Japan, UK and USA) countries with the Prophet model.  
Moreover, it allows comparisons with the RMSE (root mean square error) statistic to evaluate the performance 
of the analysis results. G7 countries were chosen because data sharing problems are less for these countries, 
and doubts about the accuracy of the number of cases announced are minimal. Based on the results obtained 
with the Prophet model, the results closest to reality with the lowest RMSE value were obtained first for 
Canada and then for Germany. Likewise, it was found that the prediction values for the first day generally 
have lower RMSE. In other words, RMSEs increase as we move away from the actual data for prediction.  
The next section of the study presents the literature review and the theoretical framework for the model. In 
the third section, empirical findings will be presented. The last section of the study will provide a general 
evaluation of the research and analysis.  

2. Literature Review and Theoretical Background  

2.1. Literature Review  

It is explained the results of SIR, SEIR, SEIRU, SIRD, SLIAR, ARIMA, ARIMA and SIDARTHE models 
used in the prediction of the spread mechanism, peak and decline of Covid-19 cases and the difficulties in 
creating predictions. The results of studies conducted with these models for different countries are graphed, 
and the performance of the models are compared with actual values and deviation value of predictions. 
Regarding this issue, it was shown that the highest deviation was found in the simple mathematical model and 
SIRD model for California.  
The study used ARIMA model to forecast the trend of the COVID-19 outbreak in Italy, Spain, and France, 
which are the countries that were most affected by the pandemic in Europe, using data from the period of 
February 21 to April 15, 2020. The different past period ARIMA models were compared with the MAPE 
performance value and the ARIMA (0,2,1), ARIMA (1,2,0), and ARIMA (0,2,1) models were selected for the 
countries, respectively. With these selected models, short-term predictions were made for the period of April 
16 to April 25, 2020. The study is showing that the ARIMA models can be used to effectively predict the trend 
of the COVID-19 outbreak, which can help governments and healthcare providers to prepare better and 
allocate resources more efficiently.  
It is aimed to obtain forecasts for two days, namely February 11 and 12, using an ARIMA model with the 
Covid-19 case counts for the period between January 20 and February 10, 2020, published by Johns Hopkins 
University. They emphasized that case definition and data collection for cases should be simultaneous to obtain 
more realistic predictions.  
It is aimed to predict the number of positive cases of different influenza (for H1N1 and H3N2 viruses) that 
may occur in 2016 with the number of pediatric cases of the influenza season between 2007 and 2015. To this 
end, they used both ARIMA and seasonal ARIMA, that is, SARIMA (an ARIMA model that can capture 
seasonal effects-seasonal autoregressive integrated moving average). The prediction results of the models are 
evaluated according to performance criteria. Accordingly, it is shown that the ARIMA model gives better and 
more realistic results than the SARIMA model in case prediction. 
Time series models are used (ARIMA and SARIMA) and a machine learning model (Prophet model). The 
prediction is based on daily and cumulative Covid-19 data for the United States, India, and Brazil, and they 



Yeliz Zoğal (2022) 
 

11 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

obtain short-term prediction results. Specifically, the Prophet model, which can capture periodic features in 
the data, gives better results for the US forecasts, while the ARIMA model gives better results for Brazil and 
India, whose cumulative cases tend to grow. The SARIMA model also captures daily cases' seasonal 
characteristics and provides better prediction results 
It is used ARIMA from time series models and Prophet, GLMNet (generalized linear model elastic net), 
random forest and XGBoost from machine learning models to predict Covid-19 cases. Based on the results of 
the analysis, ARIMA and Prophet models are more appropriate and ideal forecasts for the countries in 
question, especially the ARIMA model gives better results in Afghanistan, Bangladesh, India, Maldives, and 
Sri Lanka. They explained that the random forest machine learning model was excluded due to its poor fit to 
the data set 
It is obtained short-term prediction results with Covid-19 data for India for the period January 30 - December 
7, 2020. The results were obtained using ARIMA and some machine learning models such as Prophet, LSTM 
(long short-term memory), RNN (recurrent neural network), GRU  
(Gated recurrent unit) and LSTM-GRU models. R2 and RMSE values were obtained for numerical 
comparison of the performance of the models. In conclusion, it is shown that the LSTM-GRU model is 
superior to the others with high R2 and low RMSE values [15].  
It is used Covid-19 data for twenty countries and obtained short and long-run predictions using SEIR, 
polynomial regression, ARIMA and Prophet model. According to the prediction results, the polynomial 
regression model gives the best short-term predictions, while the SEIR model gives the best long-term 
predictions  
It is developed a new exponential growth model in their study, arguing that improving epidemic models can 
be more helpful in explaining different periods of an epidemic. The model aims to characterize the stage of 
the epidemic, especially when it shows an upward trend, and to capture the changing epidemic profile for that 
period. To this end, the model is applied to eight different infectious diseases with various transmission routes 
in twenty different geographies and the same infectious disease (Ebola) in different periods. The results show 
that the growth rate of the same infectious disease changes over time and how different geographical and 
social conditions affect the growth rate. It is explained that the epidemic growth rate is primarily influenced 
by limited population contact structure, behavioural change over time or early control interventions. 
In a study, the authors have highlighted the failures of models used to predict the spread of infectious diseases. 
The failure of models used to predict the spread of COVID-19, has made this situation even more pronounced. 
The reasons for this failure are poor data input, poor modeling, inaccurate and inconsistent assumptions, the 
predictors being overly sensitive, the distinctive features of the outbreak not yet fully determined and included 
in the models, the lack of accuracy of existing prevention measures, lack of transparency of data, lack of 
determining parameters, and reporting errors. However, solutions for some of these issues have been proposed 
such as making wave predictions instead of point predictions and selecting models that are developed and 
expanded based on performance results 
In a study, the authors aimed to use various time series forecasting models such as Prophet, Holt-Winters, 
LSTM, ARIMA, and ARIMA-NARNN to predict short-term daily and cumulative case forecasts of Covid-19, 
model the general trend of the outbreak, and model the time series based on linear and non-linear features. 
The results obtained were compared with various statistical measurements. In this regard, it was reported that 
the models showed good performance, but the ARIMA and NARNN (ARIMA-NARNN) hybrid combination 
had the best performance 
In a study, the aim is to estimate the extent of the COVID-19 outbreak in Pakistan and case forecast predictions 
using ARIMA, Diffusion, SIRD and Prophet Models. The short-term forecast results obtained show 
similarities and indicated that the highest number of infectious cases could be reached between June 2020 and 
July 2020. Due to this reason, it is conveyed that most of the population is under the threat of COVID-19 and 
that the measures taken by the government should be reviewed and improved 

2.2. Theoretical Background  

Prophet model is a time series forecasting model developed by Facebook's Core Data Science team in 2018. 
It is designed to make forecasting future data points as simple as possible and is particularly well-suited for 
business time series data. Prophet is a procedure for forecasting time series data based on an additive model 



Yeliz Zoğal (2022) 
 

12 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects [13]. The model 
has three different components  

  𝑦�(𝑡�) = 𝑔�(𝑡�) + 𝑠�(𝑡�) + ℎ(𝑡�) + 𝜀�𝑡�  (1) 

Here, 𝑦�(𝑡�) is the number of cases, 𝑔�(𝑡�) is the trend, that is, the trend function that captures non-periodic 
changes in the time series, 𝑠�(𝑡�) is the seasonality, that is, the part of the time series that captures periodic 
(weekly or yearly) changes, ℎ(𝑡�) is the part that represents holidays that occur at irregular intervals, and 𝜀�𝑡� is 
the error term that includes specific changes that cannot be covered or handled by the model [17]. The 
functions handled in the model are as follows: 

   𝑔�(𝑡�) = 1 +𝑒�𝑥�𝑝�(−𝐶�𝑘�(𝑡�−𝑚�))   (2) 

As the growth in Facebook is similar to growth in a natural ecosystem (just like the increase in cases), a logistic 
growth model was added to capture the increase in the trend. C denotes the carrying capacity, k denotes the 
growth rate, and m denotes the offset parameter.  

 𝑛�𝜋�𝑡�   𝑠�         (3)  

 𝑝� 𝑝� 
Time series are often susceptible to multi-period seasonality due to the human behaviour they represent. This 
can be a five-day working week, vacation schedules or school holidays that follow each year [17]. To capture 
these periodic effects, the Fourier model is used. Here p is the periodic term the time series is expected to have 
regularly. 

  ℎ(𝑡�) = 𝑍�(𝑡�)𝜅�   (4)  

  

  𝑍�(𝑡�) = [𝑖�(𝑡� ∈ 𝐷�𝑖�), … , 𝑖�(𝑡� ∈ 𝐷�𝑖�)]   (5)  

New year holidays, religious holidays or events that may have a country-wide impact are predictable shocks 
for time series models. For this reason, the impact of such holidays on the model is included daily. 𝐷�𝑖� is 

defined for each holiday 𝑖� in the model. Therefore, that time series interval represents the presence of that 
holiday if it coincides with that period. The 𝜅� parameter is defined as the corresponding change parameter in 
the forecast if the day of that holiday changes in the time series.  
Since the Prophet model includes both logistic and Fourier, it captures periodic waves in the observations 
more efficiently, allowing for relatively better predictions with outliers in the data. It works best with time 
series that have strong seasonal effects and several seasons of historical data. Prophet is robust to missing data 
and shifts in the trend, and typically handles outliers well [13].   
One of the frequently used statistical values for comparing the performance of model results is the root mean 
square error (RMSE). Here, 𝑦�̂ is the prediction value, 𝑦� is the actual value, and n is the total number of 
observations. RMSE is calculated as the square root of the mean of the squared differences between predicted 
values and actual values. The formula is as follows: 

   √∑ (𝑦�̂−𝑦�)2   (6)  

𝑛� 
The RMSE value is expressed in the same units as the original data, so it can be directly interpreted in terms 
of the problem being solved. The smaller the RMSE value, the better the model is at predicting the actual 
values. However, it should be noted that comparing RMSE values between different datasets or problems can 
be misleading as the scale of the data and the specific problem objectives can be different.  

3. Empirical Findings  

In the study, the daily number of Covid-19 cumulative cases for G7 countries was obtained from Our World 
in Data for 180 days, starting with the day of the first occurrence. The date range for each country varies 
according to the day the case first started. In the prediction analyses for each country, the forecasted first day 
is the prediction for the 181st day, the forecasted second day is the prediction for the 182nd day, the forecasted 
third day is the prediction for the 183rd day, the forecasted fourth day is the prediction for the 184th day, and 
the forecasted fifth day is the prediction for the 185th day. 
Table 1. Forecast and actual values  

 
     1st day  2nd day  3rd day  4th day  5th day  

( 𝑡
 

) = � ∑ 𝑎
 
𝑛� cos 

2 𝑛�𝜋�𝑡� 
+ 𝑏

 
𝑛� sin 

2 ∞ 
𝑛� = 1 



Yeliz Zoğal (2022) 
 

13 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

 𝑦�̂ (forecasted)  112,220  112,754  113,281  113,779  114,163  

Canada  𝑦� (actual)  112,281  112,663  113,399  113,836  114,172 RMSE  

𝑦�̂ (forecasted)  

France  𝑦� (actual)  217,517 

 218,753  219,844 

 219,928  219,932 

RMSE  

𝑦�̂ (forecasted)  

Germany  𝑦� (actual)  204,964  205,269  205,609  206,242  206,926 RMSE  

𝑦�̂ (forecasted)  

Italy  𝑦� (actual)  246,776 

 247,158  247,537 

 247,832  248,070 

RMSE  

𝑦�̂ (forecasted)  

 𝑦� (actual)  25,680  26,312  27,107  28,088  28,867  

RMSE  

𝑦�̂ (forecasted)  

 USA  𝑦� (actual)  3,828,431 3,896,716 3,964,163 4,031,940 4,106,884  

RMSE  445.20  40.55  458.62  1,103.05 2,528.40  

For G7 countries, the predictions of the number of cumulative cases for the 181st, 182nd, 183rd, 184th, and 
185th day, the actual number of cumulative cases of these days, and RMSEs for each day are reported in Table 
1. For France, Germany, Italy, and the UK, the RMSE increases with distance from the period used for the 
prediction between day one and day five, while for Canada, Japan, and the USA, the RMSE varies. When 
analyzed in detail, the estimated number of coincidences for Canada for the five days ranged from 112,200 to 
114,163. The actual number of cases during this period ranges from 112,281 to 114,172. The lowest RMSE 
value for Canada was obtained on the fifth day. For France, predicted values ranged from 222,122 to 232,624, 
and actual values ranged from 217,517 to 219,932. The lowest RMSE value for France was obtained on the 
first day and increased until the fifth day. For Germany, the estimated number of cases ranged from 204,765 
to 205,939, while the actual cases ranged from 204,964 to 206,926. The lowest RMSE for Germany was 
obtained on the first day. For Italy, the predicted cumulative cases were 246,860 on the first day and 250,159 
on the last day (the fifth day). The cumulative number of cases for the five days varies between 246,776 and 
248,070. The lowest RMSE for Italy was obtained on the first day. The estimated cumulative number of cases 
for Japan varies between 25,877 and 28,961. The cumulative number of cases ranges from 25,680 on the first 
day to 28,867 on the fifth day. The lowest RMSE for Japan was obtained on the fourth day. For the UK, the 
predictions for the number of cases ranged from 301,982 to 307,628 for the five days. The actual number of 
cases ranged from 301,455 to 304,685. The lowest RMSE for the UK is obtained for the first day of the 
prediction. Finally, the estimated cumulative data in the USA are 3,834,404 for the first day and 4,072,962 for 
the fifth day. The actual cumulative cases during this period vary between 3,708,557 and 3,708,557.  
The lowest RMSE for the USA was obtained on the second day.  

Japan  RMSE       

UK  

𝑦�̂ (forecasted)  

𝑦� (actual)  301,455  302,301  303,181  303,942  304,685  

4.58   6.78   8.77   4.23   0.69   

222,122   224,749   227,016   228,708   232,624   

343.20   446.94   534.56   654.42   946.02   

204,765   205,044   205,280   205,604   205,939   

14.83   16.77   24.52   47.55   73.57   

246,860   247,480   248,205   249,144   250,159   

6.25   24.00   49.81   97.79   155.71   

25,877   26,562   27,300   28,103   28,961   

14.71   18.63   14.39   1.09   6.97   

301,982   303,226   304,626   306,014   307,628   

39.30   68.92   107.73   154.46   219.38   

3,834,404   3,897,260   3,958,010   4,017,141   4,072,962   



Yeliz Zoğal (2022) 
 

14 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

The lower and upper bound predictions of the estimated numbers of cumulative cases according to the prophet 
model are reported in Table 2. For Canada, except for the fifth day, the predicted number of cases on all other 
days fell between the lower and upper bounds. For France, the number of cases on the fourth and fifth days 
was below the lower predicted value. In Germany, the number of cumulative cases remained within the 
predicted range. The number of cumulative cases in Italy on the first day only, in Japan on the fourth day, and 
in the USA on the second day only was within the predicted range. For the UK, the number of cumulative 
cases on all days was below the lower bound of the prediction.  

Table 2. Interval of forecasted value  

       1st day  2nd day  3rd day  4th day  5th day  

3,837,258 3,900,328 3,961,460 4,021,477 4,077,922  

  

The graphs of actual and predicted predictions are presented in Figure 1. It can be seen that the predicted 
values fit well with the actual values. The RMSE values also support this.  

  

Canada  

France  

Germany  

Italy  

Japan  

UK  

USA  

Forecasted  
𝑦�̂ 𝑙�𝑜�𝑤�𝑒�𝑟�  

𝑦�̂ 𝑢�𝑝�𝑝�𝑒�𝑟�  

Forecasted  

𝑦�̂ 𝑙�𝑜�𝑤�𝑒�𝑟�  

𝑦�̂ 𝑢�𝑝�𝑝�𝑒�𝑟�  

Forecasted  
𝑦�̂ 𝑙�𝑜�𝑤�𝑒�𝑟�  

𝑦�̂ 𝑢�𝑝�𝑝�𝑒�𝑟�  

Forecasted  
𝑦�̂ 𝑙�𝑜�𝑤�𝑒�𝑟�  

𝑦�̂ 𝑢�𝑝�𝑝�𝑒�𝑟�  

Forecasted  
𝑦�̂ 𝑙�𝑜�𝑤�𝑒�𝑟�  

𝑦�̂ 𝑢�𝑝�𝑝�𝑒�𝑟�  

Forecasted  
𝑦�̂ 𝑙�𝑜�𝑤�𝑒�𝑟�  

𝑦�̂ 𝑢�𝑝�𝑝�𝑒�𝑟�  

Forecasted  
𝑦�̂ 𝑙�𝑜�𝑤�𝑒�𝑟�  

𝑦�̂ 𝑢�𝑝�𝑝�𝑒�𝑟�  

112,112   112,640   113,158   113,637   113,979   

112,309   112,871   113,401   113,930   114,373   

214,108   215,628   218,392   220,465   224,020   

230,679   233,821   235,715   236,962   240,967   

204,238   204,441   204,610   204,761   204,764   

205,315   205,611   205,950   206,579   207,308   

246,557   247,171   247,832   248,778   249,692   

247,153   247,788   248,568   249,534   250,632   

25,823   26,507   27,241   28,037   28,893   

25,931   26,621   27,363   28,164   29,032   

301,638   302,846   304,237   305,568   307,094   

302,349   303,605   305,058   306,505   308,225   

3,831,657   3,894,363   3,954,755   4,013,203   4,067,776   



Yeliz Zoğal (2022) 
 

15 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

 
Figure 1. Cumulative graph and forecasted trend of countries  

4. Conclusions  

The rapid spread of the Covid-19 pandemic among countries has prompted the need for research on infectious 
diseases’ spread mechanism and predictions. This study aims to analyze Covid19 five-day prediction forecasts 
for G7 countries using the Prophet model, one of the machine learning models. For this purpose, five-day 
forecasts, five-day forecast intervals and RMSE statistics for the results obtained with the Prophet model were 
calculated.  
Based on the forecasting results obtained using the Prophet model, Canada's closest predictions (lowest 
RMSE) were obtained. Canada and Germany, Italy, Japan, Italy, Japan and the UK have particularly close 
predictions for the first day. The possible reasons for the close predictions are that these countries have 
implemented nationwide shutdowns and did not change their data enough to affect the results during the 180 
days. The most distant predictions (highest RMSE) are obtained for the USA. The reason for the highest 
RMSE for the USA is the parameter added to the model for shutdown days. In the USA, closure decisions 
were made at the state and county level rather than the federal government at the beginning of the pandemic 
and were inconsistent. For France, which has the highest RMSE results in predictions with the USA, the reason 
can be attributed to data corrections at the beginning of the pandemic. In light of these results, it is observed 



Yeliz Zoğal (2022) 
 

16 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

that the prophet model as a machine learning model provides accurate predictions when the correct data is 
provided, the number of data-related corrections is reduced, and a regular shutdown regime is followed.  
Forecasting the spread of COVID-19 is important for policymakers because it helps them to make informed 
decisions about how to respond to the pandemic. Accurate forecasts of the number of cases can inform 
decisions about lockdowns, school closures, and other public health measures. They can also help 
policymakers to plan for the distribution of vaccines and other medical resources. Additionally, forecasts can 
help policymakers to identify areas of the population that may be at particularly high risk, so that they can 
target interventions to those groups. Overall, the policy makers would consider the prophet model in 
forecasting the spread of infectious to limit the adverse effects on economies and businesses.  
Funding: This research received no external funding.  
Acknowledgments: The author is a PhD student at Institute of Graduate Programs, Ankara Haci Bayram Veli 
University. This article is partly derived from the PhD thesis and extended with additional analyses.   
Conflicts of Interest: The author declares no conflict of interest.  

References  

Anirudh, A. (2020). Mathematical modeling and the transmission dynamics in predicting the Covid-19 - What 
next in combating the pandemic. Infectious Disease Modelling. 5:366-74.  

Baldwin, R. (2020). The supply side matters: Guns versus butter, COVID-style. Retrieved from 
https://cepr.org/voxeu/columns/supply-side-matters-guns-versus-butter-covid-style  

Benvenuto, D., Marta G., Lazzaro V., Silvia A. and Massimo C., (2020). Application of the ARIMA Model on 
the COVID-2019 Epidemic Dataset. Data in Brief. 29:105340. doi: 10.1016/j.dib.2020.105340.  

Ceylan, Z. (2020). Estimation of COVID-19 Prevalence in Italy, Spain, and France. Science of The Total 
Environment. 729:138817. doi: 10.1016/j.scitotenv.2020.138817.  

Duman, A. (2020). COVID-19 ile Artan Eşitsizlikler ve Yoksulluk. Retrieved from 
https://sarkac.org/2020/06/covid19-ile-artanesitsizlikler-ve-yoksulluk/  

Furtado, P., (2021). Epidemiology SIR with Regression, Arima, and Prophet in Forecasting Covid-19. The 7th 
International conference on Time Series and Forecasting. MDPI.  

He, Z. and Tao, H. (2018). Epidemiology and ARIMA Model of Positive-Rate of Influenza Viruses among 
Children in Wuhan, China: A Nine-Year Retrospective Study. International Journal of Infectious 
Diseases. 74:61-70. doi: 10.1016/j.ijid.2018.07.003.  

IMF. (2021). World Economic Outlook.  

IMF. (2022). World Economic Outlook.  

Ioannidis, J. P. A., Sally C., and Martin A. T. (2022). Forecasting for COVID-19 Has Failed. International 
Journal of Forecasting. 38(2):423-38. doi: 10.1016/j.ijforecast.2020.08.004.  

Malik, M. I. (2020). Analysis and Forecast of COVID-19 Pandemic in Pakistan. medRxiv. doi: 
10.1101/2020.06.24.20138800.  

Prajapati, S., Swaraj, A., Lalwani, R., Narwal, A. and Verma, K. (2021). Comparison of Traditional and Hybrid 
Time Series Models for Forecasting COVID-19 Cases. arXiv. doi: 
https://doi.org/10.48550/arXiv.2105.03266.  

Robson, W. (2019). The Math of Prophet. Retrieved from https://medium.com/future-vision/the-math-of-
prophet-46864fa9c55a.  



Yeliz Zoğal (2022) 
 

17 
American Interdisciplinary Journal of Business and Economics | 

https://sadipub.com/Journals/index.php/aijbe 

 

Roubini, N. (2020). A Greater Depression?. Retrieved from https://www.project-
syndicate.org/commentary/coronavirusgreater-great-depression-by-nouriel-roubini-2020-03  

Sah, S., Surendiran, B., Dhanalakshmi, R., Mohanty, S. N., Alenezi, F. and Polat, K. (2022). Forecasting 
COVID-19 Pandemic Using Prophet, ARIMA, and Hybrid Stacked LSTM-GRU Models in India. 
Computational and Mathematical Methods in Medicine. 2022:1  

Sardar, I., Muhammad A. A., Víctor L., Ahmed A. and Pradeep M. (2022). Machine Learning and Automatic 
ARIMA/Prophet Models-Based Forecasting of COVID-19: Methodology, Evaluation, and Case Study 
in SAARC Countries. Stochastic Environmental Research and Risk Assessment. doi: 10.1007/s00477-
022-02307-x.  

Taylor, S. J. and Letham, B. (2018). Forecasting at Scale. The American Statistician 72(1):37-45. doi: 
10.1080/00031305.2017.1380080.  

Viboud, C., Lone S., and Chowell, G. (2016). A Generalized-Growth Model to Characterize the Early 
Ascending Phase of Infectious Disease Outbreaks. Epidemics. 15:27-37. doi: 
10.1016/j.epidem.2016.01.002.  

Voyvoda, E., and Yeldan, E. (2020). Salgın, Türkiye Ekonomisi ve Gerçekçi Bir Kamu Politikası Önerisi. 
Retrieved from https://sarkac.org/2020/06/salgin-turkiye-ekonomisi-ve-gercekci-bir-kamu-politikasi-
onerisi/  

Wang, Y., Zehui Y., Wang, D., Yang, M., Li, Z., Gong, X., Wu, D., Zhai, L., Zhang, W. and Wang, Y. (2022). 
Prediction and Analysis of COVID-19 Daily New Cases and Cumulative Cases: Times Series 
Forecasting and Machine Learning Models. BMC Infectious  

WHO. (2020). WHO Director-General’s opening remarks at the media briefing on COVID-19.  Retrieved 
from https://www.who.int/director-general/speeches/detail/who-director-general-s-opening-remarks-
at-the-media-briefing-oncovid-19-11-march-2020  

World Bank. (2020). Projected Poverty Impacts of COVID-19 (coronavirus). 


