Lecture 11: Data Analytics and Decision Making

Question 1

Outline key predictive analytics techniques

Answer 1

Predictive analytics is the art of obtaining information from collected data and utilising it for predicting behavioural patterns and trends.
- Business knowledge + Data mining = Predictive analytics

Main types of predictive analytics are:
- Basic analytics, involving the slicing and dicing of data into smaller sets and data visualisation; basic monitoring of large volume of data; and also identification of anomalies
- Advanced analytics, using algorithms for complex analysis of either structured or unstructured data such as predictive modelling and text & voice analytics etc.
- Operationalised analytics, which is when analytics become part of the business process. E.g. insurance company building model that predicts the likelihood of fraudulent claims.
- Monetising analytics, using analytics to increases revenue from their own operations, recognising that datasets may be valuable to other companies. E.g. customer data of credit-card providers, financial institutions etc. – ‘Data is the new currency’
Ethical issues involved (compliance, privacy and security issues limiting ways in which data can be used)

Question 2

Tutorial 12 Q1:

Five broadly applicable ways to leverage Big Data

Answer 2

1. Creating transparency
   • Making big data more easily accessible to relevant stakeholders in timely manner
   • “Open data” principles in which more raw government databases are made publicly available; linking and information sharing of databases across business or government department to cut costs, improve efficiency and effectiveness and reduce the ‘silo’ effect
2. Enabling experimentation to discover needs, expose variability, and improve performance
   • As organisations create and store more transactional data in digital form, they can collect more accurate and detailed performance data on everything from product inventories to employee performance data
3. Segmenting population to customise actions
   • Big data allows organisations to create highly specific segmentations and to tailor products and/or services precisely to meet those needs
   • E.g. online retailers such as Amazon, Netflix, and use of loyalty cards to target individuals with customised promotions and advertising.
4. Replacing/supporting human decision making with automated algorithms
   • Automated stock trading;
   • Fraud risk analyses in many settings such as: purchasing and invoicing; fraudulent insurance claims; identifying potential rogue trading; making credit risk assessments and collectability assessment
5. Innovating new business models, products and services
   • New ‘disruptive’ business models e.g. Uber, Air BnB, AirTasker, Netflix, Kaggle

Question 3

Tutorial 12 Q2:
“Most people are not good intuitive statisticians.” Evaluate statement in accounting context. What is the normative decision model that should be used? Why is it used or not used in practice?

Answer 3

• The above statement is correct, both in everyday life and in business contexts as well.
• A decision is a choice between future alternative courses of action. Therefore, many decisions do not have objectively ‘correct’ answer since they are based in subjective probabilities of future events.
• Examples include a bank loan officer deciding whether to extend credit to a customer, and if so at what interest rate; a manager deciding on whether to invest in future project; and an investor assessing the future probability of firms in which they have invested.
• The normative decision model of probabilistic judgment that should be used in these scenarios is Bayes’ Theorem. The theorem states that:

Posterior odds (revised probability) = Likelihood ratio (amount by which prior expectation should be revised) X Prior odds (initial probability or base rate)

• There is a vast amount of research evidence that individuals do not in fact apply Bayes’ Theorem in making judgements in real life. We are not good intuitive statisticians.
• Evidence shows that base rate information is either under- or over-weighted and new info may be under-weighted.
• This occurs because humans have limited cognitive processing capacity and when we have to process increasingly large amounts of info, our brains often take ‘shortcuts’ to cope with this complex judgement tasks simplified by use of heuristics and biases.

Question 4

Tutorial 12 Q3:
Explain implications for accounting if decision makers in an accounting context display any or all of the representativeness, availability or anchoring and adjustment heuristics.

Answer 4

• Heuristics may be efficient and effective when used in a professional context; e.g. the use of 5-10% thresholds for materiality judgements in accounting (when we need some concrete guide to a subjective matter)
• However, they may reduce quality in judgments and decisions if they result in significant biases.

Three main types of heuristics:

1. Representativeness, where items or events that are viewed as more representative in the population is assessed as having higher probability of occurrence than those less representative.
   - Representativeness clearly essential in sampling, which is used extensively in many business decisions.
   - The problem is that it can result in stereotyping, in base rate info being under-weighted/ignored, and other decision-relevant information being under-weighted/ignored. (jumping into wrong conclusions)
   - E.g. auditors’ samples, fraud judgements, judgements of loan default/corporate failure
2. Availability, where the assessed probability of event is based on the ease at which instances of the event come to mind.
   - Leads to ignoring of base-rate probabilities. Instead, sensational events or those that are heavily covered in the media, are overestimated.
   - In accounting context, this can lead to misallocation of scarce resources, e.g. ignoring small everyday risks that have a large cumulative impact, compared to spending to prevent a sensational but highly unlikely event.
3. Anchoring and Adjustment, where individuals are too strongly attached/ ‘anchored’ to an initial belief, response, number or action. Thus they insufficiently adjust for new information, i.e. new information is under-weighted.
   - E.g. not changing internal controls to respond to changed risks; not changing last year’s work program to respond to changed risks; not adjusting sufficiently for new info in the use of fair value or asset impairment modelling.

Question 5

How can we reduce negative effects of heuristics and biases in decision making?

Answer 5

• Business knowledge (experience, intuition, people skills) + rational modelling = improved quality of judgment
• Increased education and training
• More use of algorithms and other decision support models
• Changing format and presentation of information to improve data visualisation
• Working in teams?