We have the following problem:
At the beginning of every year, a gardener classifies his soil based on its quality: it's either good, mediocre or bad. Assume that the classification of the soil has a stochastic nature which only depends on last year's classification and never improves. We have the following information:
- If the soil is good, then the classification next year will be good ($p=0.2$), mediocre ($p=0.5$) or bad ($p=0.3$).
- If the soil is mediocre, then the classification next year will be mediocre ($p=0.5$) or bad ($p=0.5$).
- If the soil is bad, then the classification next year will be bad ($p=1$).
The gardener can also pay to have the soil fertilized, which will change the probabilities as follows:
- If the soil is good, then the classification next year will be good ($p=0.3$), mediocre ($p=0.6$) or bad ($p=0.1$).
- If the soil is mediocre, then the classification next year will be good ($p=0.2$), mediocre ($p=0.6$) or bad ($p=0.2$).
- If the soil is bad, then the classification next year will be good ($p=0.1$), mediocre ($p=0.4$) or bad ($p=0.5$).
The costs for fertilization are $\$1000$ for good soil, $\$2000$ for mediocre soil and $\$10,000$ for bad soil.
The gardener can sell his soil: good soil sells for $\$10000$, mediocre soil sells for $\$6000$ and bad soil sells for $\$2000$.
When should the gardener fertilize, and when shouldn't he?
I want to tackle this problem using Markov chains but I'm having trouble. I know the state space is given by $S = \{ \mbox{good, mediocre, bad} \}$, but I'm having trouble formulating the Markov chains. Would appreciate help.
Soil condition at start of the year is known, and fertilization will only change the soil next year, so in taking a decision, we should look at its effect on next year's profits.
If soil is "good", next year's profits (thousands of dollars) will be:
$0.2*10 + 0.5*6 + 0.3*2 =5.6$ without fertilization,
and $0.3*10 + 0.6*6 + 0.1*2 - 1 = 5.8 $ with fertilization,
so fertilize
Work out similarly for the other 2 states.
Note that you will have to take such decisions every year.