Data Warehouse 1.0 - Common Query Examples From Standard Reports
Introduction
Mifos Data Warehouse 1.0 is a star schema data warehouse. Its purpose is to make reporting on data as easy and as available as possible. This can mean a number of things like:
- maximise the ease of writing SQL queries or using query products on top of the data structure
- an increased confidence in data quality resulting from the data analysis that's part of moving from the large mifos data structure to the small star schema data structure
- reports running faster
The idea is to increase the access and availabity of data over time. One part of this is to continue to simplify the data structure over time by doing extra work upfront during data loading and by the use of meta data to hide difficulties.
This document isn't a tutorial on how to write sql or exactly what's in the data warehouse but it should give a flavour for writing queries against the data warehouse and the best place to get examples (the standard reports shipped with mifos). Contains:
- Examples of vanilla but important reporting uses
- The way to get at loan arrears information quickly and easily
- Examples of customer and account based reporting that is not so vanilla
Vanilla But Important Reporting on the Star Schema
The following query shows how to get a breakdown of repayment transactions for a branch for a quarter. This is a typical star schema type query. The 'fact' table is joined to the 'dimension' tables that are used for filtering and grouping.
Repayment transactions cover normal repayments, adjustments, reversals, reschedules and write-offs.
select aa.account_action_name, sum(r.principal_amount + r.interest_amount) as amount
from fact_loan_repayments r
join dim_account_action aa on aa.account_action_key = r.account_action_key
join dim_date d on d.date_key = r.action_date_key
join dim_office o on o.office_key = r.branch_key
where d.year_quarter = '2010-Q1'
and o.display_name = 'Branch No. 54'
group by aa.account_action_name
order by amount desc
This simple query does quite a lot. Its a trivial matter to change 'branch' for 'loan officer', 'center', 'group' and other tables that are used for filtering and grouping data. There are other 'fact' tables in the data warehouse that cover such things as disbursals and savings. It's the same approach to querying for all of them (consistency of query pattern is a strength of this type of data warehouse).
Finally, here's a query to get a breakdown of the number and amount of disbursals per loan_officer by branch by month for 2009 (loan officers can move branch).
select p.display_name as loan_officer, o.display_name as branch, d.month_name, sum(disbursal_count) as disbursals, sum(disbursal_amount) as disbursal_amount
from fact_loan_disbursals disb
join dim_date d on d.date_key = disb.disbursal_date_key
join dim_office o on o.office_key = disb.branch_key
join dim_personnel p on p.personnel_key = disb.loan_officer_key
where d.year4 = 2009
and p.is_loan_officer is true
group by p.display_name, o.display_name, d.month_number
order by p.display_name, o.display_name, d.month_number
Loan Arrears Information
Because it's quite complicated, the data warehouse pre-calculates information about each loan that is in arrears. Data is snapshot by day. The query below gets arrears information that is past a mifos configured 'lateness' number of days as of end 2010-03-31. It does it for a loan officer and breaks the results down into pre-determined weekly bands. Replacing 'weekly' by 'monthly' give the same data in monthly bands.
select wb.past_due_description,
count(distinct(hla.customer_key)) as clients, count(distinct(hla.loan_account_key)) as loans,
sum(total_in_arrears) as total_in_arrears,sum(total_outstanding) as total_outstanding
from hist_loan_arrears hla
join dim_date endperiod on endperiod.date_key = hla.as_of_date_key
join dim_arrears_band_weekly wb on wb.arrears_band_weekly_key = hla.arrears_band_weekly_key
join dim_personnel lo on lo.personnel_key = hla.loan_officer_key
join dw_mfi_configuration config on config.mfi_configuration_key = 1
where endperiod.date_value = date('2010-03-31')
and lo.display_name = 'XXX'
and hla.days_in_arrears > config.loan_lateness_days
group by wb.past_due_description
order by wb.arrears_band_weekly_key
Customer and Account Based Reporting
In the mifos standard reports... there's a little of the vanilla reporting above, quite a bit of the loan arrears reporting but mostly it's customer and account based. This type of reporting against the data warehouse is a little more involved than the vanilla but there are 2 main positives.
- it's still better than querying the mifos application database
- most importantly, it caters for customer hierarchy changes (loan officer movements, group and client transfers etc) as well as customer and account status changes. Basically,historical reporting like comparing one time period to another is made possible.
The main tables (dim_customer, dim_loan and dim_savings) each link directly to a full mifos hierarchy... branch, loan officer, center, group etc... but the thing which is different from normal mifos tables is that one customer, for example a client, will have a number of entries in dim_customer. Each time the hierarchy changes (e.g. client or group transfer, loan officer assignment) or there is a status change... a new entry is added. The entry is valid for a period of time which is denoted by a pair of date fields called valid_from and valid_to. If you want to find the unique version of the client that is relevent on 2010-01-01 you add the following to a query...
and valid_from <= date('2010-01-01')
and valid_to > date('2010-01-01')
It took me a little while to get used to this. Also some of the queries felt a little complicated but still better than working with the mifos application database structure. Below are a couple of examples from the standard mifos reports. If you have the Pentaho Report Designer you can get the .prpt files from git://mifos.git.sourceforge.net / mifos/bi and look at the queries.
a) as of 2010-01-01, how many centers did loan officer XXX manager
select count(distinct(customer_id)) as centers_managed
from dim_personnel p
join dim_customer c on c.loan_officer_key = p.personnel_key
where p.display_name = 'XXX' /* get loan office entry that is valid for date*/
and p.valid_from <= date('2010-01-01')
and p.valid_to > date('2010-01-01')
and c.customer_status = 'CenterStatus-Active'
and c.valid_from <= date('2010-01-01') /* get center entry that is valid for date*/
and c.valid_to > date('2010-01-01')
b) as of 2010-01-01, how many clients with active loans did loan officer XXX manage
select count(distinct(c.customer_id)) as clients_with_loans
from dim_personnel p
join dim_customer c on c.loan_officer_key = p.personnel_key
join dim_loan l on l.customer_key = c.customer_key
where p.display_name = 'XXX'
and p.valid_from <= date('2010-01-01')
and p.valid_to > date('2010-01-01')
and c.customer_level_id = 1 /* client... without this groups would get picked up as well*/
and l.loan_status in ('AccountState-ActiveInGoodStanding', 'AccountState-ActiveInBadStanding')
and l.valid_from <= date('2010-01-01')
and l.valid_to > date('2010-01-01')
c) as of 2010-01-01, what were the repayments on active loans that loan officer XXX dealt with. Here's where I struggle a bit. It's a little struggle technically but really its a bigger struggle to make sure you are asking the right question. The query below only looks at non-adjusted repayments for active loans managed by a loan office at a certain point in time... which is okay if that's what the question is.
select ifnull(sum(principal_amount),0.0) as principal_paid, ifnull(sum(interest_amount),0.0) as interest_paid
from fact_loan_repayments paid
join dim_loan l on l.loan_account_key = paid.loan_account_key
join dim_personnel p on p.personnel_key = paid.loan_officer_key
join dim_account_action aa on aa.account_action_key = paid.account_action_key
where p.personnel_id = 'XXX'
and p.valid_from <= date('2010-01-01')
and p.valid_to > date('2010-01-01')
and paid.action_date <= date('2010-01-01')
and aa.account_action_id = 1
and paid.adjusted is false
and l.loan_status in ('AccountState-ActiveInGoodStanding', 'AccountState-ActiveInBadStanding')
and l.valid_from <= date('2010-01-01')
and l.valid_to > date('2010-01-01')