*Bounty: 50*

*Bounty: 50*

I’ve an online customer data which has the purchases made in every month and recency of the purchases information for 12 months. So data looks like below:

```
SUB_ID, month, spend, last_spend(days), spend_frequency
```

I’m using the Marov Chain model to estimate the customer lifetime value for the fix period (with time horizon). For the same, I’m computing the Transition Probability Matrix (TPM) from the subscriber data with taking into account the transitions are making in one period. Then I’m using the formula for CLV with finite time horizon as below. Reference of the formula is here

Where, P is the TPM with period 1 and R is the Revenue vector in the first period.

My problem is this method is giving me very high error (in the order of ‘000 when it comes to percentage error).

Please help if I’m doing anything incorrectly here.

Note: I’ve considered the outlier elimination. However that’s not helping much.