Real Time Bidding

Real Time Bidding


Real Time Bidding (RTB) is a method of displaying online ads in real time. Due to the strict time limit (usually less than 120 ms) the method is reserved only for automated systems using highly specialized algorithms and efficient computers. Usually, the company’s profit is related to the activity performed by the user after viewing the advertisement on the website. Displaying the advertising without any interaction with the user means a loss for the company. Therefore, the mathematical models are created in order to select the users who are most likely to perform an activity (e.g., click on an ad, fill in the form, subscribe to the newsletter).

RTB is a model example of what Big Data issues entail. By definition, the Big Data is characterized by:

  • large amount of data (currently at least millions of objects),
  • large variety of data,
  • variability, i.e. high speed of data receiving.

For the reasons mentioned above, it is not possible to distinguish the users manually. Dedicated, automatic traffic filters are the only solution to the problem of cheaters detecting.

Displaying one advertising is usually cheap (counted in micro dollars), but the number of such advertising is counted in tens of billions a day. Thus, without a control mechanism, the company could spend the entire budget in a few minutes. However, using the predictive algorithms, one can ensure a cost control in the RTB system related to the online bidding traffic. Hourly spending limits are determined based on historical data.

This approach allows to spread the costs throughout the day instead of spending them at the beginning of the day. In order to improve the accuracy, the proposed method considers the panel econometric model in which hours are the panels. The results are evaluated on the basis of comparative tests between the panel model (fixed effects model) and the time series model (least squares estimator). It turns out that in most cases the panel method gives more accurate forecasts ordinary.

Percentage of costs per hour for an example day with two forecasts indicated: a panel fixed effects estimator (solid line) and a ordinary least squares method (dashed line). Source: Bernardelli M. (2017)