Comparative analysis of estimated market values of companies: Forbes 2000 and BIST 388 company example

companies. The Istanbul Stock Exchange, which has been serving more than 300 companies with a volume of billions of dollars since 1985, offers various opportunities to international investors. Of course, investors have to apply to many analyzes of companies and companies in order to evaluate these opportunities correctly and to make a profit. There-fore, it is of great importance for millions of investors around the world that firm and market data are highly transparent, consistent and predictable. Because otherwise, markets that cannot show stability in developing countries cause the market values of the company to rise and fall excessively with the opportunities and crises that constantly occur. This situation leads to unnoticeable manipulations in firm and market values and investors to be mistaken due to the inability to analyze them correctly. The purpose of this research is to estimate the market values of 2019 Borsa Istanbul firms by using Artificial Neural Networks (ANN) method with the data of Forbes 2000 companies in 2019. As a result of the analysis, it was revealed that the companies in Borsa Istanbul have higher or lower market values than they should be, and the results are also supported by the comparisons of ratio analysis.


Introduction
Equity markets are a public market in which company shares are traded at an agreed price and offer investors the opportunity to share the company's profits based on the amount of ownership (Barnes, 2016). Investors need to analyze the market and the company correctly in order to predict these ups and downs. However, manipulations that may arise with high competition can cause competition and mislead investors (Aggarwal and Wu, 2003). Future price predictions of stocks can also be made by methods such as artificial neural networks (Khan etc., 2011). In addition, there are many studies showings that the values of stocks offered for sale during the global crisis decreased to lower levels than expected (Boudriga and Ghachem, 2016). In this risky conjuncture, one of the most concrete methods of providing information about the business performance and economic variables of companies is ratio analysis (Drake and Fabozzi, 2012). However, for these analyzes to yield clear results, company data should be compared with companies in global markets and unreliable markets or companies should be distinguished from others. Ratio analysis is one of the main indicators by which investors can predict the financial performance of companies they want to buy stocks (Nikolai etc. 2009). In these analyzes, variables such as the sales amount, profit, assets and market values of the companies are used.

Profitability of actives
The profitability ratio of assets is used to evaluate to what extent a firm's ability to make a profit with respect to its total assets (Berman et al., 2008). It also specifies how it uses assets to generate profit for efficient management. Return on assets is calculated by dividing the annual earnings of a firm by total assets (Angenieux, 1964). It can be said that the higher the coefficient of the result, the more successful the companies are in making a profit.

Price sales ratio
The price selling ratio is found by proportioning the total market value to the sales. This ratio shows how much the investor pays for every $ 1 profit the company generates (Brigham & Houston, 2012). It is better for investment decisions to have a small coefficient of the ratio indicating how many times the firm value is the business volume. In order to analyze how high or low this ratio is, firstly, an average ratio should be determined by making a sectoral comparison (Giacomino & Mielke, 1993).

Share price measurement
The high number of coefficients of the share price measurement calculated by the ratio of firm value to profit indicates that the share price is high, and the low number of times indicates that the price is low (Shım & Sıegel, 1988). It shows how many times the net profit is on an annual basis within the market value of the company (Taner & Akkaya, 2004). This ratio is generally preferred to be low (Bhagat & Black, 2001). At the same time, whether the ratio is low or high can be better understood by comparing it with the Price / Earnings ratio of other businesses operating in the same sector or with the average price earning ratio of the sector. F / K (Share price measurement) = Market Value / Net Profit for the Period

Literature review
The study in which Singh and Srivastava used Deep Learning (DL) for stock data prediction can be considered as one of the first analyzes on this subject, and it was revealed that the deep learning method gave better results than other neural networks (Singh & Srivastava, 2017). In a study conducted on NSE Stock Market, 2 deep learning models were used for stock prediction and it was concluded that the deep learning method performed better than linear time series models (Hiransha etc., 2018). In a study conducted on the China Stock Exchange in 2018, a deep architecture-based model was built for stock market prediction and the use of artificial neural networks in analysis with big data was proven to be an effective method (Chen etc., 2018).
Vargas, Lima, and Evsukoff have tried to predict the movements of the Standard & Poor index with the Deep Learning method based on the news published in financial articles, and have conducted a study that provides evidence that even words can be converted into financial predictions (Vargas etc., 2017). In a study conducted for Borsa İstanbul, the estimation of stocks was made with the help of Artificial Neural Networks and it was proved that the performances of ANN models are around 70% more efficient than other models (Kara etc., 2011). Research by Dase & Pawar revealed that Artificial Neural Networks (ANN) methods are useful for predicting the stock index and also have a feature that can be used to predict whether it is best to buy, hold or sell stock market shares (Dase & Pawar, 2010 ).
In the study of Yetis, Kaplan and Jamshidi, ANN model was used for the estimation of stock market prices and as a result of the research; It has been stated that for individual and institutional investors, financial analysts and financial news users, anticipating the future behavior and movement of stock prices can help them to show the necessary behaviors in advance in order to gain more profit or not to lose (Yetis et al., 2014). In an application by Qiu, Song and Akagi on the Japanese Stock Exchange, estimates were made using ANN and it was concluded that not only market value but also real import and export values, discounts, interest and bond estimates could be estimated (Qiu et al., 2014). Cao, Leggio and Schniederjans made a comparison between Fama and French's model and artificial neural networks to predict the Chinese stock market, and given the decreasing availability of information in emerging markets and the sometimes questionable quality of information, investors' It has been found to be more usable to increase predictive power (Cao et al., 2005).
Zhang and Wu applied the ANN model for the estimation of the S&P 500 index, and with different optimizations integrated into the artificial neural network, better estimates were achieved with less computational complexity in stock market forecasts (Zhang & Wu, 2009). Patel and Yalamalle analyzed a number of stocks in the Indian Stock Exchange and it was found that the ANN technique was very useful in predicting stock prices of a particular company as well as stock indices (Patel & Yalamalle, 2014).

Material and method
In this study, many options of the Phyton program were used to analyze the company data in Forbes 2000 according to the characteristics of the companies in Borsa Istanbul and determine their market values. In the study, the entire data set consisting of information about 2000 companies should be analyzed and weighted in each Turkish company. In other words, for the market values entered into the program, it should compare all the data and choose the most approximate market value. Since the data set is slightly larger than the standard survey data and 2 thousand analyzes are required each time, high data analysis speed is required. Therefore, such processes may require machine learning and statistics to work together (Atalay & Çelik, 2017): In the study, Artificial Neural Networks (ANN) method was used as one of the artificial intelligence applications.

Artificial neural networks (ANN)
In information technology, ANN is a hardware and / or software system modeled after the study of neurons in the human brain. ANNs are also called neural networks for short and are various deep learning technologies that fall under the umbrella of artificial intelligence or machine learning (Rouse, 2019). Neural networks method is particularly successful in detecting nonlinear patterns. Artificial neural networks, pattern recognition and optimization are also used. Artificial neural network applications include supervised and unsupervised learning (Çelik, 2018: 122). In its simplest form, the neural network is a "sensor" consisting of a single neuron. Much like biological neurons with dendrites and axons, a single artificial neuron is a simple tree structure with input nodes and a single exit node connected to each input node (Willems, 2019). The comparison of biological neuron and ANN is given below.
Biological Neuron versus Artificial Neural Networks Input nodes: Each input node is associated with a numeric value that can be any real number. Real numbers make up the full spectrum: they can be positive or negative, integers or decimals.
2. Links: Each link leaving the login node has a weight associated with it, and this can also be any real number.
3. Then all the values of the input nodes and the weights of the links are combined together: they are used as inputs for the weighted sum: or expressed in a different way, 4. This result will be the input for a transfer or activation function. In the simplest but trivial case, this transfer function will be an identity function. For sure, the output may not be a linear function: it may be a discontinuous function. Since this can cause problems in mathematical processing, the sigmoid function, which is a continuous variable, is often used. As you already know, a sigmoid function is also a logistics function. Using the logistics function provides a much simpler result.
5. As a result, you have the exit node associated with the function of the weighted sum of input nodes (such as the sigmoid function). The sigmoid function is a mathematical function that results in an "S" shaped curve mathematically as follows: 6. Finally, the sensor can give with an input node advancing to 1 permanently, a parameter called bias. The deviation value is critical because it allows you to scroll to this. Left or right, I can determine the success of your learning.
ANN usually includes a large number of processors running in parallel and arranged in layers. The first stage receives raw input information similar to the optic nerves in human visual processing. Each successive layer receives output from the layer preceding it rather than the raw input, so neurons further away from the optic nerve receive signals from those close to it. The last stage produces the output of the system. Each compute node has its own small area of information, including the rules it sees and is initially programmed or developed for itself. The layers are highly interconnected; This means that each node in layer n will be connected to multiple nodes in layer n -1 (inputs) and layer n +1 providing input data for those nodes. The output layer can have one or more nodes where the output it produces can be read (Rouse, 2019). Artificial neural networks stand out for their adaptability, which means they change themselves as they learn from initial training, and subsequent studies provide more information about the world. The most basic learning model focuses on weighting input streams; it is how each node weighs the importance of the input data from each of its entries. Inputs contributing to obtaining correct answers are heavier (Rouse, 2019).
Deep learning is a subfield of machine learning, a set of algorithms inspired by the structure and function of the human brain. These algorithms are often called Artificial Neural Networks (ANN). Deep learning is one of the hottest areas in data science, with many case studies in the fields of robotics, image recognition, and Artificial Intelligence (AI) with surprising results. One of the most powerful and easy-to-use Python libraries for developing and evaluating deep learning models is Keras; It complements the efficient numerical computing libraries Theano and TensorFlow. The advantage of this is that it basically enables us to get started with neural networks in an easy and fun way (Willems, 2019).
Deep learning networks are separated in depth from the more common single-hidden layer neural networks; that is, the number of node layers that the data must pass through in a multi-step pattern recognition process. Previous versions of neural networks like the first sensors are shallow, consisting of an input and an output layer, and at most one hidden layer in between. More than three layers (including input and output) qualify as "deep" learning. In deep learning networks, each node layer is trained on a different set of features based on the output of the previous layer. The further you go into the neural network, the more complex the features your nodes can recognize, as they collect and recombine features from the previous layer (Pathmind, 2020).

Application
In this study, analyzes were made using the "TensorFlow", "Keras" and "Scikit-learn" libraries of the Phyton program. In addition to these, "Pandas", "Numpy", "Seaborn and" Matplotlib "libraries were also used. Tensor-Flow in Python is among the main platforms used in the research and development of Deep Learning. Creating deep learning models with TensorFlow is quite difficult. For this reason, Keras Python library is used to create a deep learning model on TensorFlow. Keras is a minimalist Python library for deep learning that can run on Ten-sorFlow or Theano. Keras was developed to make the application of deep learning models for research and development as quick and easy as possible (Brownlee, 2019). Scikit-learn is a free library that provides many unsupervised and supervised learning algorithms in Python. It also has various algorithms such as the Scikit-learn support vector machine, random forests, and k-neighbors.
In that research, the possible market values of Borsa Istanbul firms are estimated by comparing the data of Forbes 2000 firms with the values of the firms in Borsa Istanbul. In the analysis, continuous value estimation from ANN types, one of the machine learning methods, was used. By this way, according to the sales, profit, assets and market value data of the Forbes 2000 company, the estimated market value was reached by entering the sales, profit and asset values of the companies entered from Borsa Istanbul. The data of the Forbes 2000 companies for the years 2019-2018 are available from the website https://www.someka.net/excel-template/forbes-global-2000-list/, and the data of 389 companies in Borsa Istanbul are available to the Public Disclosure Platform. (KAP) has been accessed from the site https://www.kap.org.tr. The number of layers, outputs and parameters formed as a result of the Artificial Neural Network are given as in Table 1. Considering Table 1, since the number of data is high, it is seen that 1 Input layer has 4 Hidden layers and 1 Output layer. The high number of hidden layers in complex data provides better results. The summary of 2019 Forbes 2000 data, which was first entered into the Python program in the research, is shown in Table 2. At the same time, 2018 asset data were entered into the program in order to compare asset turnover ratios.  Descriptive statistics of Forbes 2000 data are given in Table 3. The summary of Borsa İstanbul company data and ratio analysis included in the Python program in the 2. stage of the research is given in Table 4. Descriptive statistics of Borsa İstanbul data are shown in Table 5. (Except for ratio ratios, data are $ million times.)  The income, profit, asset and market values of the companies included in Forbes 2000 were entered into the program, and then the income, profit, assets of the Turkish company data were entered one by one and the market value was requested from the program. At the same time, ratio calculations were made in order to measure this uncertainty and Table 6 includes the estimated and actual values and ratio ratios of Borsa Istanbul companies. The ratio of period net profit to total assets used in profitability analysis shows how profitable companies are according to their total assets. In the graphs comparing Turkish companies and Forbes 2000 companies, changes in the period coefficients have been observed and these differences are shown in Figure 4. Although there are companies that use their investments efficiently from companies in Borsa Istanbul, their inefficiency rates appear to be significantly higher than Forbes companies. In line with the data obtained, the reason for the lower investment returns of Turkish companies seems to be the inability to generate profit. When this situation in Turkish companies is revised, companies can be in a position to earn more with less investment. The column chart version of the asset profitability differences of Borsa İstanbul companies and Forbes 2000 companies is given in figure 5. The price / sales ratio indicating how many times the company value is the business volume and how much money companies will pay for each unit of sales are found. When the graphs of Turkish companies and companies in Forbes 2000 are examined, it is seen that the ratio of Forbes companies is smaller. If this ratio is small, it indicates a positive situation but a high does not mean that the company is cheap. However, when Borsa İstanbul companies are analyzed by sector, it is concluded that some companies are smaller than their multiplier averages. These differences are shown in Figure 6. It has been concluded that companies with low price / sales ratio, which is very important for investors, make future pricing or company shares are quite expensive. With future pricing, which is a risky situation, companies promise that they will make higher profits in the future than their current profits. In this case, it is necessary to pay attention to whether there is manipulation or not and the relevant institutions should follow this situation. The column chart of the price-selling ratio differences between Borsa Istanbul companies and Forbes 2000 companies is given in Figure 7. Turnover ratio of assets (actives) is a kind of risk measure. The ratio is obtained by the formula of proportioning annual sales to total assets. As it can be understood from the formula, increasing the turnover of the company or decreasing the balance sheet will increase the asset turnover rate. It can be said that there are idle resources in companies that have the effect of reducing the turnover rate. The higher turnover of the companies listed in Forbes 2000 in terms of million $ in the charts than the companies in Borsa Istanbul indicates that the companies have less risk factor. These differences are shown in Figure 8. Although the ratio varies from sector to sector, the fact that this ratio is 1.5 and above indicates that the company is doing enough. It can be said that the situation of companies with 2 and above is quite good. Based on the graphics, it can be said that companies in Borsa Istanbul have lower asset turnover rates than Forbes 2000 companies. The column chart of the asset turnover differences between Borsa İstanbul companies and Forbes 2000 companies is given in Figure 9. The analysis performed is within the scope of fundamental analysis and is based on estimating the value of a stock or the entire market. The point distribution graph of the share price measurement ratio differences of Borsa İstanbul companies and Forbes 2000 companies is given in Figure 10. Share price measurement is one of the financial ratios frequently used by both shareholders and managers. The fact that the rates of Turkish companies are higher than they should be is a proof that there are directions within the company and they can be easily affected by possible exchange rate fluctuations. This will create a particularly negative situation for small investors. Investors will continue to suffer as long as they are not examined by the necessary institutions and penal sanctions are imposed. The column chart of the share price measurement ratio differences of Borsa İstanbul firms and Forbes 2000 firms is given in Figure 11. A sectoral analysis reveals that the real market values of the Factoring and Banking sectors are quite high and have a rate-increasing effect. Among the sectors, the companies that use their assets in the best way have been mining / quarry and chemical product enterprises. While it is a good situation to have companies with high ratios, having low sectors can be a dangerous situation. It was observed that the Factoring and Construction companies made a more profitable investment on a sectoral basis, from the rate at which the companies were calculated how much money would be paid for each unit of sales. Figure 12 shows real market value, active profitability in Figure 13, and sectoral differences of price sales ratios in Figure 14. It is seen that the difference between the actual value created by the Turkish companies in Borsa Istanbul and the estimated value emerging in the software is quite high. As a result of the analysis, this situation is acceptable and these ratios that give clues about the real and estimated market values of the companies are given in Figure  17. When the possible reasons for this difference seen in the chart are investigated, it can be concluded that the company shares are at a very high level of what they should be, the asset management is not carried out well and the shares of the company that are traded in the stock exchange are exposed to manipulative effect. In addition, it is an inevitable end that the company shares in this situation will fall sharply in possible crises. In order to prevent this situation, the control over the stocks should increase and the win-lose situation should be replaced by a winwin policy. Column chart showing the differences between actual and estimated values is given in Figure 18. When the correlation table of the ratio analysis is examined, the relationship between the real market value and the stock price measurement is at a moderate level, while the price selling ratio is at a high level close to the level of perfection, the relationship between the other variables and the real market value is weak or 0. On the other hand, there is an excellent relationship between the price selling ratio and the share price measurement plot and they are highly affected by each other. If the asset is profitability, there are 3 variables and it is weakly related to variables such as profit / loss, estimated market value and asset turnover. The ratio in 2019 sales is moderate with the estimated market value, while there is a zero or no relationship with 2018 assets and 2019 profit / loss with other weak variables. In the case of profit and loss, there is a high relationship with the estimated market value, a moderate relationship with 2019 and 2018 assets, while there is a weak relationship between return on assets and 2019 sales. The strongest relationship in predicted market value is affecting the profit / loss situation of 2019, 2019 sales, 2019 assets and 2018 assets moderately affecting this value. The assets that affect 2019 the most are 2018 assets at the level of excellence with 0.99. The only variable in asset profitability can be mentioned, and this is the weak relationship between 2019 sales. The correlation graph is shown in Figure 19. Figure 19. Correlation matrix

Conclusion
In this research, the data of the best 2000 companies in the world according to Forbes were taken as optimum and the estimated market values of Turkish companies in Borsa Istanbul were obtained. According to the research, the market value of 235 of 388 Turkish firms was higher than the estimated market value, while the market value of 153 firms was below the estimated values. To support these results and to understand where the problem lies, ratio analyzes were made and compared with Forbes 2000 companies and Borsa İstanbul companies.
The first thing that stands out in ratio analysis is the asset transfer results. According to Forbes 2000 data, none of the Borsa Istanbul companies is above 1.5. In this case, it can be said that Borsa İstanbul companies do not use their assets at 100% capacity or that the idle capacity is high. It is seen that the value of textile and mining enterprises among these companies is high. Likewise, the company with the closest value to 1.5 is SÖNMEZ textile company. In order to understand whether the investments made by the companies in general are correct or not, the return on assets ratios, which show how efficiently the companies are invested according to their total assets, have been determined.
It was determined that 209 of the companies were above the general average of 0.03 according to their return on assets. However, these data should be based on the industry average. Only 209 of Borsa Istanbul companies are above the general average, and 179 are below the general average. Sectorally, mining is the most productive investment sector. In the analysis of both the asset turnover rate and the return on assets ratio, it is seen that some Borsa Istanbul companies keep this amount as investment amount instead of distributing it as dividend, although they make a profit. However, it can be said that this investment, which is spent instead of dividends, is not efficient enough due to idle capacities.
There are 2 other ratio analyzes to get information about market values, namely price to sale ratio and share price measurement. According to share price measurements, 380 companies are below the general average. According to the price-sales ratio, 381 companies are above the general average. These ratios should be lower than the industry average, not the general. The clearest result that can be said as a result of share price measurement and price sales ratio analysis is that banks and factoring companies make excessive profits according to their firm values. Because there is little need for fixed assets in both sectors, it can keep assets and firm value lower than profits.
Borsa Istanbul is the only market in which Turkey has shares. The Turkish economy, which is important for foreign investors in terms of markets, can be affected by small fluctuations in exchange rates and Borsa Istanbul indices can be affected either extremely negatively or extremely positively. According to the analysis, although these firms have high potential, they do not seem profitable enough. For this reason, Borsa İstanbul companies should be well supervised by state authorities and investors should be able to trade on more transparent data.
In the study, the ratio can be calculated on the basis of company data and in general, but researchers who want to work on this subject can reach comparative information about companies and markets by reaching ratio averages on a sectoral basis or they can reach stronger results, for example, they can detail the research by accessing equity information. Valuable investors who will invest in Borsa Istanbul are recommended to invest using the methods shown in the research, to examine the balance sheets, cash flow statements and income statements of companies, and to invest by taking into account the country's economy. In this way, both the information about how to invest in a company and the extent to which the company will be affected by possible manipulative movements will be known in advance.

Author contribution statements
Authors contributed equally to the design and implementation of the research, to the analysis of the results and to the writing of the manuscript.

Disclosure statement
No potential conflict of interest was reported by the authors.

Ethics committee approval
Ethics Committee Approval is not required for this study. All responsibility belongs to the researchers.