A framework for solving problems better - understand what goes into the numbers

"Garbage in, garbage out"

As we all know, you can only get out what you put in, so the quality of data inputs drives the quality of outputs for decision making.

One of the things I I like to apply in problem solving is situation-specific understanding and simple rules of thumb to maximise quality such as:

1. Data inputs should relate to a representative time period and sample
2. The effort required in reconciling the data inputs at both a micro and macro level should be proportional to the importance of the end decision
3. Any data limitations should be communicated to and understood by all participants

Let's use a topical case-study to illustrate.

Case-study: managing UK inflation

The Bank of England is tasked with using interest rates and monetary policy to deliver a target 2% inflation outcome for CPI (consumer prices index).

So how is CPI calculated? In simple terms the index is the value of a "shopping basket" of items with their price indices weighted by relative spend value.

The data inputs used for price indices and weights ultimately drive calculated CPI, government policy and the interest rates paid by consumers and businesses.

Let's look further using our situation-specific thinking:

Price indices

ONS consumer price indices are compiled using a monthly sample of over 700 goods and services that have their price movements measured in approximately 20,000 outlets within the UK.

1. Representative? Given that we all purchase different items from different places 700 seems a small dataset
2. Reconciled? Comparing calculated price indices with known price point changes could only improve accuracy. Data at scale from key merchants on major spend items would be an obvious start.
3. Limitations understood? Detailed information is provided by the Office for National Statistics (ONS) but a simple guide always helps. In fairness the ONS folk are very helpful in providing clarification and additional information on request.

Weights

Data sources are between one and two years old at time of writing.

1. Representative? Given the significant recent price increases we have all experienced (many well above the headline rate of inflation) does using a dataset this old give the accuracy needed?
2. Reconciled? Deriving comparison weights from actual payment data could improve accuracy significantly
3. Limitations understood? Benchmarking CPI data with other data sources could enable informed debate on accuracy and suggestions for improvement.

Recent Banking sector volatility demonstrates how every basis point counts.

Wage negotiations, costs and contract re-pricing are driven by similar calculations for other inflation metrics such as RPI (retail price index).

These material impacts demonstrate the importance of making calculations as accurate as possible.

What's your view on how much we understand about CPI?