In this paper, it is proposed to use the Google Trends service for the identification of interest in risk control trends through queries in different languages of users in the world and Ukraine for such risk control terms: "risk control", "forensic", "financial investigations" in different languages. The time series formation model was described and analyzed, taking into account the high asymmetry and seasonality of the development of forensic in the regions of Ukraine and the world. Google Trends have sufficient opportunities to determine users' level of interest in major trends in the financial investigations, which allows forming a statistical base for decision making for modeling and further development of this direction. The experimental modeling showed the seasonality for "forensic" and "risk control" popularity trends as well as the determination of the eastern and western in Ukraine in the relation of "forensic".
Crisis phenomena in the world, the imperfection of the market economy have created favorable
conditions for the distribution and activation of economic criminality, all more companies bear loss in
the different spheres of their activity. The amount of financial machinations puts under a doubt the
efficiency of threat estimation and implemented technologies for protection against economic crimes,
as well as the application of the necessary measures to counteraction of fraud. Thus, according to
research of the World Economic Crimes Review for 2020, approximately every second organization
in the world (47%) for the last two years has faced various types of fraud. This is the second most
important level of economic crime in the last 20 years, losses from fraud in fixed cases from 99
countries amounted to more than 42 billion dollars USA [
Risk control systems are created to activate the spread of the forensic process. To confirm the actualization of such an opinion in society, a search for approaches to the possible definition of patterns of interest by the theme of financial risk control has began, analysis of the formed interest in this theme, opinions and public confidence in determination the essence of the linguistic and terminological list in this direction.
In modern conditions, domestic and foreign search engines not only accumulate and group
information, provided by modern search tools. The researches of search query statistics have recently
become widespread and special actuality [
At the same time, to prevent economic risks, it is important to use Google's tools for determination
interest in financial risk control through the introduction of forensic as a method of investigation
intra-corporate cases of fraud and factoring risk management [
There are no separate researches that analyze the capabilities of the Google Trends service to analyze users demand in search engines for linguistic terminology.
Thus, the presented experience of using Google Trends is progressive and actual to the possibility of solving the problem of analysis the level of users interest in the world and in particular in Ukraine, the topic of financial risk control.
The contribution of the research is the result of the Google Trends interest data processing using time series decomposition. It allows to establish the popularity development patterns and find the synchronous 52 weeks seasonality interest to the search of the "risk control" and "forensic" terms. At the same time the interest in searching for "forensic" nearly 10 times larger. The analysis of interest in other definitions ("financial investigations") and different geographic locations confirmed that the popularity level is stable but without significant seasonality.
Google Trends is a service that allows to identify the interests of the target audience, analyse
demand for goods or services, track current and past trends, popular events, as well as find out the
number of Google queries from different regions for certain keywords [
Google Trends provides statistical data about search requests performed in Google at a certain
period with grouping by countries, categories, and search types (web, images, news, YouTube) [
A review of scientific papers proves that Google Trends is an effective tool not only for market
research and collection of statistical information, but also for forecasting demand and making final
decisions in business [
The efficiency of the Google Trends tools use by interested users of information on financial risk control has shown a growing interest in it. The strategic direction is to ensure greater access to open data, as there is a prospect of developing standards for open data in open information networks.
When planning areas of research in the field of financial risk control, a key term is defined "forensic", with which the terms "risk control" and financial investigations are connected. Each query is monitored as the dynamics of popularity of these queries in Ukraine and worldwide grows over the past 5 years. For a more detailed analysis, the resulting data is exported in CSV format, which contains data tables.
In addition to statistics, search results were obtained across regions, which allow identifying
strategic geographical centers that are actively promoting and developing the innovation sector in the
field of finance. The results of the "forensic "search query in Ukrainian among the regions of Ukraine
are shown in Figure 1[
In order to explain and analyze such asymmetry in using a generic term "forensic" in technologies of risk control in finance in relation to the interest in Ukraine, investigated the frequency of the term "risk control" through the Google Trends search and analysis system, which was considered in various language submissions: Ukrainian, Russian, Polish, English. The results of the comparison revealed that the most popular query English −"risk control" (Figure 2), while there are insufficient data for processing and analysis in Ukrainian − "risk contro"l"р(изик-контроль"), in comparison with in Russian "risk control" ("риск-контроль") and in Polish "risk control" ("kontrola rysyka"), where the frequency of queries in English exceeds several times (Figure 3).
Analyzing the results from Figure 3, there is the some interest and additional comparative analysis can be conducted in the beginning of 2018, when there was a strong increase of interest in risk control in Russian and Polish at the same time.
Examining the results of the search for all countries "risk control" in English, was identified countries that are interested in this topic with the possibility of generating statistics of queries for analysis: most interested − Botswa nima,baZbwe, Ethiopia, Zambia, Tanzania, Australia, Singapore, least − France, Mexico, Brazil, Turkey, Japan, Poland and Russia. "RiskEncgolinsthrorle"mainins the prevailing term in the world.
According to the investigation, interest in the term "risk control" can be traced worldwide, in particular in North America, Western Europe and Africa. Analyzing the data for 2020, it should be noted a persistent trend to the increase of interest in the term "risk control" in English (Figure 2). The Table 1 shows the analysis results of the requests frequency of selected terms in different languages in the world and in Ukraine for 2016-2020.
As can be seen from Table 1, the tendency to search for the term "risk control" in English does not coincide worldwide and Ukraine in terms of Latin search for forensic, this indicates that forensic is not identified only with risk control, but rather seen as a means of crisis management, so as forensic in Africa and Asia.
However, the number of results on the queries "financial investigations", "risk control", "financial investigations" In Ukrainian and Russian, especially in Ukraine, is too small, that even after reducing the analysis period − there is not enough anadlyattiacssubfomrission, so there was a transition to finding interest in the concepts forensic and financial investigations. 5. Estimation the search investigations" terms frequency of "forensic" and "financial
The study of the request frequency for the term "forensic" was considered in language submissions: forensic in Ukrainian, forensic in Russian, forensic in English. The most popular query is forensic in English, and the activity of the query was localized in such countries: Zimbabwe, Singapore, South Africa, UAE, Jamaica, India, Ghana, Qatar, Kenya, Nigeria, USA, Canada, UK, Australia, Sri Lanka. If considered within Ukraine, it is Vinnytsia, Kharkiv, Kyiv, Odesa, Dnipro, Lviv, Kherson and Donetsk regions, there is also a concentration of both western and eastern regions and the lack of interest of the center. It is shown in Figure 4 and Figure 5the request frequency for the term "forensic" in English, "forensic" in Ukrainian and "forensic" in Russian worldwide.
As can be seen from Figure 4, there is a persistent trend when searching for the term "forensic" in English, while queries for the term in Ukrainian and in Russian (Figure 5) have clearly defined peak values and seasonality. The same situation is observed in the analysis of the request frequency in Ukraine (Figure 6). Analysis of the popularity of inquiries in Ukraine and worldwide over the past year confirms the data obtained on the peak and seasonality.
As can be seen from Figure 6, the general interest in the term "forensic" in English in Ukraine is constant, meanwhile in Ukrainian and in Russian only in some peak periods, which requires additional measurements.
The study of the request frequency of the term "financial investigations" was considered in the language representations: "financial investigations", "financial investigations" in Ukrainian ("фінансові розслідування"), "financial investigations" in Russian ("финансроавсыселедования"), and "financial investigations" in Polish ("dochodzenia finansowe"). As can be seen from Table 1, only in English and in Russian there is available information for analysis.
Analyzing Figure 7, it is possible to mark that by the requests in the world's leads "financial investigations" in English and "financial investigations" in Russian. In relation to requests in Russian, peak values are similarly traced in some periods, as in previous requests on forensic in both Russian and Ukrainian. Requests by the term "financial investigations" can be traced from the countries, such as the United States and the United Kingdom". Thus, there is no such diversity in the world as in term of "risk control". Analysis of requests for the last year has also shown interest in this term around the world, but, it should be noted, in contrast to the term "forensic", more pronounced peak values. It is not possible to analyze the interest in this term in Ukraine due to the lack of the required amount of data. Thus, the process of developing financial investigation technologies in Ukraine is just beginning.
The obtained data prove the interest in the concepts of "risk control", "forensic" and "financial investigations" both worldwide and in Ukraine represent their orientation for development in these areas. It should be noted the existing asymmetry of interest in terms both worldwide and in the regions of Ukraine, as well as the presence of clear peak values, which became the basis of the next task to build a model of time series.
There are two ways to detect the seasonality in time series data [
There are three common components in the time series data [
Two time series formation models are typically used: (additive) and (multiplicative).Additive model is used when the variance of the time series doesn't change over different values of the time series. In our experiments we assume additive model.
The time series process consists of the steps depending on the components assumed to be present but always includes detrending and deseasonalization.
Estimation of the trend of time series is usually performed with centered moving average (w-MA) procedure, where is the width of the moving window. The calculation of trend component values * +, ̅̅̅̅̅̅̅̅̅is written as follows: ⌊ ⌋ ∑ ⌊ ⌋ where is the length of time series. The choice of parameter may come from the expected seasonality lag, but not necessarily. Additionally, if the hypothetical length of the seasonality period is sufficient, the centered MA procedure will shorten the length of the initial time series by (if is odd). So, the selection of mostly comes from practical considerations.
We propose to preserve first and last
values of the trend series with the averaging of only available values of the initial time series at the beginning and end. For example, for 7-MA ⌊ , first three values are calculates as ∑⌊ ⌋ , ∑⌊ ⌋ ⌋ . Similar procedure is applied at the end of the series. This is shown inFigure
The traditional way to estimate the seasonality is the building of the plot of autocorrelation
function (ACF), which is sometimes called correlogram. Known correlation shows the linear
relationship between two variables, autocorrelation shows the one between the signal and its lagged
version. Autocorrelation values * + may be written as [
∑ ( ̅) where ̅̅̅̅̅̅̅̅is the lag of the series, is the maximum possible lag, ̅ is the average of the series.
Making the decision on the ACF plot is based not only on absolute values of the coefficients but
also on their significance. The typical approach includes the building of the confidence bands and
analysis if the coefficients are bigger or less than band values. If data are random, ACF values should
not exceed corresponding significance band values. The widths of the confidence bands are calculated
according to [
√ where is the number of samples, is the cumulative distribution function of the standard normal distribution and is the significance level (equals to 2.58 for the 99% significance level and 1,96 for 95%).
So, if the data are statistically independent, the coefficients of the ACF should fall inside the confidence bands (in our experiments it was set up to 99%).
Let us investigate seasonality in different time series, provided by Google Trends service for different periods and geography of search queries.
The first definition we are looking at is the "forensic" term with the worldwide geography for the last 5 years (measurement interval in Google Trends is 1 week). The popularity time series, trend series (based on the 7-MA smoothing), and detrended series assuming additive time series model are shown in Figure 9.
The bar chart of the ACF coefficients and significance bands are presented in Figure 10. There are different suggestions about the choice of the maximum lag value for ACF building, we have chosen 30% of the length of series to build (lag varied from 1 to 77 weeks inclusively).
As one can see, there is a statistically significant value at lag 52 that means the seasonality pattern of 52 weeks. Besides this, there are also other confident peaks at lags 1, 15-16, 53, 64, 74 intervals.
The choice of moving average size window influences the ACF values. The smoothing level is not enough to catch significant seasonality patterns for small window sizes like 3-MA (popularity series and corresponding trend series are shown in Figure 11). The ACF plot after this smoothing and detrending is shown in Figure 12, there we can see a lot of peaks with a small amplitude which are under the significant bands.
The bigger allows to extract more exact long-going seasonality and more significant absolute correlation values (about 0.5). The smoothed with 15-MA trend line is shown in Figure 13. ACF plot shows that there is a clear 52 weeks seasonality interval (Figure 14).
So, the choice of the depends on the form and length of the series, presence of noise in it and practical experience. In all experiments, we used such window size that is not big enough to make the trend almost linear but at the same time allows us to analyze ACF values confidently.
The ACF plot for the "forensic" term interest in Ukraine for the last 5 years is shown in Figure 15. The corresponding original series and trend (based on the 7-MA) are shown in Appendix. There are no statistically significant results in 99% confidence bands but three of them are in 95% confidence intervals (lags 21, 42, 49). Increasing the MA window size (15, 21, 31 values were tested) change neither the common situation nor the amplitude of the ACF coefficients, so with the 99% confidence, there is no seasonality in this time series.
-0,05 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 Lag
The results of the processing of interest in "forensic" term for the last 1 year all over the world are shown in Figure 16 and Appendix. The size of moving averaging windows was set up to 7. There are no statistically significant results in 99% confidence bands for different but 6 weeks seasonality detected for 95% for 7-MA (correlation value is 0.34). It seems there is no strong seasonality here. 0,25 0,15 0,05 -0,15 -0,25
The results of the processing of popularity time series for the "forensic" term in Ukraine during the last 1 year are shown in Appendix and Figure 17Series for the "forensic" term for the last 5 years in Ukraine. There are no statistically significant results in 99% confidence bands but 12 weeks seasonality is found with 95% confidence still with a small absolute correlation value. These results are stable for different .
Looking at the interest in the "risk control" term for the last 5 years and 1 year worldwide we obtained the following results (15-MA), which are shown in Appendix and Figure 18Series for the "forensic" term for the last 5 years in Ukraine. One can see clear 52 weeks seasonality here and corresponding correlation values become stronger for greater .
Looking at the "risk control" term for only last year (Figure 19) we can conclude that the popularity is independent, the only ACF value at lag 5 gets outside of 95% bandwidth for 7-MA, and the same repeats for different .
The last term we are looking at is the "financial investigations" all over the world for the last 5 years (Figure 20) and last year(Figure 21), initial data and corresponding trend series are presented in Appendix. Three ACF coefficients are greater than the 95% confidence interval (lags 28, 40, 62), there are no values greater than 99% confidence. Increasing the MA window size doesn't change the situation for both these time series, so the interest in the search of these definitions seems to be stable but without seasonality.
As it has been shown earlier, the popularity of both "forensic" and "risk control" terms for the last 5 years worldwide has 52 weeks of seasonality. It worth noting that the interest in these term searching is synchronous (correlation coefficient is 0.61) besides the fact that the popularity of the "forensic" is nearly 10 times larger. The corresponding trends smoothed with 15-MA for the popularity of "forensic", "risk control" and scaled "risk control" time series are shown in Figure 22.
The research has shown that Google Trends has ample opportunity to determine the level of user interest in the main trends of the financial risk control. It is noted the low interest of society in the term "financial investigations", which indicates negative trends. Among searchers, the concepts of "forensic" and "risk control" are not identified, which is more seen as a means of crisis management. There is also a persistent trend to the increase of interest in the term "risk control" in English worldwide, in particular in North America, Western Europe and Africa in 2020. Some interest and additional comparative analysis can be conducted in the beginning of 2018, when there was a strong increase of interest in "risk control" in Russian and Polish at the same time. The significant predominance of the search for the term "risk control" in English in Ukraine and worldwide requires an additional study. The same situation is observed when analyzing the requests frequency in Ukraine. The analysis of the requests popularity in Ukraine and worldwide over the past year confirms the data obtained about peak and seasonality. The analysis of the requests of the term "financial investigations" over the past year has also shown interest worldwide, but, it should be noted, unlike the term "forensic", more pronounced peak values. It is not possible to analyze the interest in this term in Ukraine due to the lack of the required amount of data. So, the process of development of financial investigation technologies in Ukraine only begins.
The analysis of the popularity time series for the seasonality allowed to find52 weeks seasons for the "forensic" and "risk control" terms in the scope of last 5 years worldwide. The interest in these definitions is also somewhat synchronous in time. There is no confident seasonality for the interest in other terms being investigated.
The advantage of Google search services is the ability to measure interest in regions and cities, which allows to identify strategic geographic centers that actively promote and develop the innovation sector in the field of financial risk control, as evidenced by a certain asymmetry west-east interest in forensic and needs to search of measures to equalize the development of financial risk control throughout Ukraine, in particular, the center.
The obtained data prove the interest in the concepts of "risk control", "forensic" and "financial investigations" both worldwide and in Ukraine representing their orientation for development in these areas. It should be noted the existing asymmetry of interest in terms both worldwide and in the regions of Ukraine, as well as the presence of clear peak values, which became the basis of the next task to build a model of time series.
It is difficult to define the affinity of countries by geographical location, which needs additional analysis. It determines the direction of further research –the estimation of the relationship between regions of the world and the greatest interest in the topic of risk control. Other interesting topic for future research relates to analysis and comparison of search statistics provided by competitors for Google Trends. 9. References