





MORE INFORMATION ABOUT OUR FOOTBALL PREDICTING TECHNOLOGY
Being a team of software developers who were engaged in betting & livebetting software and who generally works for the welfare of a betting places, we were always interested in doing something for the people who visit the very same places as well. 
If the same is done for the away team, and then for all combinations of results, we'll get the values for the columns P(H) and P(A).
The most possible result is the biggest multiplication result of individual probabilities, which gives the most possible result. This is the most simple presentation of what we 'd like to show here. So, let's move on. Parameter λ which you have already met with earlier, from now on we'll call "scoring rate". Since the results of the team differ at home and away we'll introduce the terms "home scoring rate" and "away scoring rate". Now the table 1. will get a little more complicated, depending on whether the match is played at home or away for the team we want to predict the result. And still this can be calculated by the simple operations in MSExcel. This model has grown complicated even if we still haven't put exactly the same important parameter as the scoring rate, and that is the capability of one team defiance and not to receive a goal. It is true that it's the most probable that home team will gain two goals, but also the probability is the same for the guest team not to receive more then one. Of course this parameter should also be included . Even better, two parameters should be included, home defensive modifier and away defensive modifier. 
The way this formula works out the most probable result will not be described here because this is not some mathematical study. What is important is that you know two things.
1. parameters depend on each other, because it is not possible to calculate the home scoring rate without away defensive modifier known and vice versa. 2. parameters are calculated by repeating calculating process while the values of parameters are not converged to some value, i.e. that values in i+1 iteration don't differ less than, for example, 0,000001 in i iteration. If we don't use, like many authors argue, parameters that are very important: the advantage of home ground and the length of the bench, we will get a satisfying model which will lead us firmly through many games that bookmakers offer. This model was tested through all games during season in many world leagues and the results are very satisfying. It means that disposition of 12 to 14 teams (out of 18) matches exactly, while a problem exists only for the teams in the middle of the table. Since this method is based on setting of results, it means there is a great certainty one team will win, it might be a wrong result, but a winner is certain. We think there is an problem of tied games which is mentioned in some articles and we will try to overcome this problem in next versions of the application. The application in its work don't use this model. It uses a little bit complex model. The problem with previous calculations is in the fact that games at the beginning of the season has the same difficulty factor as the ones at the end. It means one should not considerate a momentary form of the team. Instead of calculating the team's parameters at once for whole season, the application goes on calculating the results again and again, each time after a match have played. Since this method is more complex from the previous, WeI won't even try to "spice" this article with formulas. They use Gamma functions and Poisson's distribution. This method is often met in literature and it's often the object of analysis. Whether someone is interested in the essence of these methods, I suggest to read some of the articles mentioned in Reference part of this article. III. HOME GROUND ADVANTAGE We all agree that in football ( and also in many other sports) a home ground is of great importance. But how this advantage could be modeled and presented in numbers. Should it be considered on a whole league level or only on a team level? Which factors should be considered? Well, we will enumerate several now, and we are sure you'll remember a group around ten yourselves. Ball possession, number of posts, number of cards, number of passes completed, number of winners of home team in a league... The database that software is using hasn't these data (it's a doubt whether other databases has these information). That's why we have let these modeling to you. Our advice to you is that it is the best, with the exception of testing, not to use this option for many reasons. You can be sure you won't guess the right value, and also the model you are using already has some information about home team advantage, through the history of the matches. IV. THE BENCH STRENGTH Maybe during weekend’s friendly conversations about games we never talk about bench strength, but we certainly do mention injuries of key players. Let's look at the Barcelona FC. The injury of Chavi has an important role in team's play in next round, but the fact that Iniesta spends his time on the bench is less important, or is to so? Software based upon the history of results simply has no information about the quality of the bench. Do you have them? Do you think it is important? Our suggestion for you is not to use this option. First, you don't even know for sure whether there would be any substitution or not, and beside this there is a fact that some of the players have, in some previous games, took part in the game, so the software will "see" them through the history of the results. If they haven't took their part until now, it is even better. It is probable they won't now. V. HEAD TO HEAD We provide you with headtohead section. Keep in mind that you need to check this option before you place a bet. You can find interesting and very usefull things here. VI. PREDICTION WIZARD OK. It's simple from now on. From the first tab of the Wizard chose country and home team, and away team. We're trying to simulate the next: Round of the season 2009/2010. Under this part of application there are 3 options, which offer the following possible choice: 1. All the competitions in which the teams took part in. We like this the most. It offers me the most games, and one can choose what he thinks is going to be useful later. 2. The competitions in which both team took part in, whether or not they had met each other. For example, if you are testing AC Milan and Manchester UTD, you’ll see that both teams took their part in the Champion league, but have not met each other yet. 3. The competitions in which both teams took part in and met each other. This should, if the teams are from the same country, be the local leagues. Let we repeat, for the above remarks numbered as 2. and 3. the number of games in the analysis is somewhat limited, and this isn't good sometimes. Since we're talking about scoring rate and defensive modifier, we cannot completely perceive condition for example, if we study only duels through long period of one season. But this choice is also up to you. Have in mind that this model (not like others) is not 100% fixed, and also it depends a lot on you and your skills. Other models just give you theirs predictions, trying to hide input parameters and limit your knowledge in calculation. There is also possibility to chose mathematical method wizard will use in it’s prediction. There are three methods at the moment, and we are always working on new once and current methods improvement. First method is based on all the games in selected seasons for calculating probabilities. Second includes current scoring rate of teams and defense modifiers, and probability is calculating after each game, so it implements current form of a team more precise. Third model is similar to second, but it makes difference between home and away matches. So don’t be surprise if you get completely different results. It’s up to you to choose appropriate option for what you have in mind, or better yet, use all models and find games that will have same results for all the models...those are the one you should go for sure.There will be many situations where all three methods will suggest different results. In this case move to the next game or go for 1X2, not for scored goals, turnovers or some other option. Our final tip is not to bet just looking at the wizard results. Use graphical presentations in “Analyze team” section of the application, look at both teams form and if you don’t find some logical result, like in previous example, just go to other game and by doing so minimize your risk. VII. CONCLUSION During time we have tested many models which deal with predicting football results. Our original idea was to gather in one place a lot of leagues and seasons results and offer them to users, and the predicting of the results was never simple. References * Predicting Football Results, Andrew James Moore May 6, 2004 * Multivariate Poisson models, Dimitris Karlis,October 2002 * Bayesian and NonBayesian Analysis of Soccer Data using Bivariate Poisson Regression Models, Dimitris Karlis and John Ntzoufras April 2003 