General instructions
This experiment comprises two subsequent stages.
In both parts, you will have to make a series of choices between investing into either a stock or bond.
For each choice, you have up to 120 seconds to decide.
If you do not make a choice after 120 second, you will automatically invest into the bond.
Contingent on your choice, you earn monetary units. The number of units you earn in every decision is summed up at the end of the experiment.
For every monetary unit you have earned, you are paid 3 Euro-Cents. In addition, you receive a participation fee of 2 Euros.
Example 1: Assume you earn 400 monetary units in the experiment. Your ultimate income in the experiment, which you are paid out, equals: 280*0.03 + 2 = 10.40 Euros.
Example 2: Assume you earn 250 monetary units in the experiment. Your ultimate income in the experiment, which you are paid out, equals: 220*0.03 + 2 = 8.60 Euros.
Please make your decisions carefully as they determine your ultimate income that you receive at the end of the experiment.
The use of any decision support which is not provided on screen is strictly forbidden.
Note: At every point, the experiment can only continue once all participants have made their decision!
Specific instructions stage 1
You participate in 7 asset market games. Each game consists of 5 consecutive rounds.
In every round, there are two assets in which you can invest. On the one hand,
there is a bond that is always paying 3 monetary units for certain. On the other hand,
there is a stock which is either paying 1 or 5 monetary units. In every round, your task is to choose whether
you want to invest into the bond or the stock.
The probability that an investment into the stock yields the high payoff (5 units),
depends on the stock's fundamental state.
With an equal probability (50%/50%), the fundamental state can either be good, or bad.
Good state: If the stock is in a good state, the probability for the high payoff equals 70%.
This is, with a chance of 70% you receive the payoff of 5, and with 30% you receive the payoff of 1.
Bad state: If the stock is in a bad state, the probability for the high payoff equals 30%.
This is, with a chance of 30% you receive the payoff of 5, and with 70% you receive the payoff of 1.
In every round, before you make your decision, you receive independent opinions of 4 experts about the stock’s payoff in the given round.
Each expectation can either say High Payoff, or Low Payoff, indicating the expert’s expectation about the stock’s payoff in this round.
The expectations can be correct or incorrect and correlate with the stock’s actual payoff in a certain way.
If you have figured out the underlying patterns, the expectations can help you identify when the stock will yield a high payoff.
In addition to the expectations of the four experts, you are shown the prediction of a
Machine Learning System that was trained to predict the stock’s payoff in a given
round based on the experts’ stated expectations.
In other words: The Machine Learning System was designed to recognize the underlying patterns.
The Machine Learning System was trained and tested on 1.000.000 distinct observations.
Each observation comprised the four expectations of the experts as independent variables and the stock’s actual payoff as dependent variable.
In a test, the trained Machine Learning System correctly predicts the stock’s payoff in a given round in 72% of the cases.
Below you can find additional information about the structure of the system. We use the Bernoulli Naive Bayes Machine Learning model.
A history of the experts’ opinions, the Machine Learning System’s prediction, and the actual payoff of the stock of previous rounds is shown in every market game.
To make sure that you have understood your task, there will be two trial rounds of decision making with feedback. Your decisions and outcomes in trial rounds do not affect your payoff.
In this trial, you have unlimited time to get familiar with the task.
NOTE: The underlying relation between expert expectations, machine prediction and stock payoff are different from those in subsequent payoff relevant rounds.
Additional Information about the Machine Learning System
The Naive Bayes Model is one of the simplest, but one of the most powerful algorithms for classification.
In the context of the experiment, classification refers to correctly predicting whether the stock will yield a high payoff or low payoff.
The Naive Bayes Model is based on Bayes’ theorem. Bayes' theorem describes the probability that an event occurs,
given prior information about conditions that might be related to the event. It serves as a way to figure out conditional probability.
In this experiment, the Naive Bayes Model uses the information from the four experts to compute the probability that the stock will yield a high payoff.
In other words, the model asks: given all the independent expectations of the four experts, what is the probability that the stock will now yield a high payoff?
Therefore, the Naive Bayes Model learned the underlying relationship between experts’ expectations and the stock’s actual payoff and can therefore
help you make the payoff maximizing choice. You are only shown a High Payoff prediction, if the probability for this event is the maximum.
Otherwise, you are shown a Low Payoff prediction.
Two examples of real world applications:
Spam Filters
Naive Bayes classifiers are a popular and effective technique for e-mail filtering.
The Naive Bayes Model works by correlating the use of words with spam and non-spam e-mails and then, using Bayes' theorem,
calculate a probability that an email is or is not spam.
Medical Diagnosis
One of the main advantages of the Naive Bayes approach which is crucial for medical diagnosis is that all
the available information is used to make a prediction. When dealing with medical data, the Naive Bayes Model
takes into account evidence from many attributes to make the final prediction. Additionally, it provides
transparent explanations of its decisions, which is why it is considered one of the most useful classifiers to support physicians' decisions.