The Co-holding Puzzle: New Evidence from Transaction-Level Data John - - PowerPoint PPT Presentation

The Co-holding Puzzle: New Evidence from Transaction-Level Data

John Gathergood1 & Arna Olafsson2

1University of Nottingham 2Copenhagen Business School & CEPR

October 2020

1/56

Overview

1 2 3

Introduction Data Results I: Measuring Co-holding

4 Results II: Determinants of Co-holding 5 Conclusion

2/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

3/56

Introduction

Individuals exhibit financial behaviours that appear inconsistent with models of rational, or quasi-rational, behaviour, e.g., Failure to refinance (Andersen et al., 2019) Choosing dominated insurance options (Bhargava et al., 2017) Making dominated debt repayment choices (Gathergood et al., 2019) Understanding such violations is important for realistic models of consumer behaviour and for policy One of the starkest violations of rationality is holding liquid high-cost debt and liquid low-yield assets simultaneously The “credit card puzzle” or “co-holding puzzle” (Morrison, 1998; Gross and Souleles, 2002)

4/56

Co-holding Puzzle

Co-holding puzzle has been documented in many settings (mostly using survey data), with a variety of explanations offered Cash liquidity management (Telyukova & Wright, 2008; Telyukova, 2013) Credit limit risk (Druedahl & Jorgensen, 2018; Gorbachev & Luengo-Prado, 2019) Within-household coordination (Bertaut et al., 2009; Vihriala, 2019) Noncognitive ability (Choi & Laschever, 2018) Self-control (Bertaut et al., 2009; Gathergood & Weber, 2014)

5/56

This Paper

We present new evidence on co-holding from transaction level data Use daily, transaction-level data from Icelandic financial aggregator which provides objective measure of financial balances Study co-holding of liquid cash deposits together with liquid bank

• verdrafts, the most common means of borrowing in Iceland

I In Iceland consumers commonly hold multiple deposit accounts as a

means of borrowing via overdrafts

I Aggregate overdraft debt is twice the value of rolling credit card debt

in Iceland, with credit cards typically used for transacting only whereas

• verdrafts are used for borrowing (Carlin et al., 2017)

This form of co-holding is i) fully liquid, ii) immediately costly, iii) costly with certainty at time of transaction

I Cleaner setting compared with studying credit cards 6/56

Results

Coholding is uncommon and typically short-lived 15% of individual × days exhibit non-zero co-holding Most co-holding is of less than 15 days worth of consumption Duration of co-holding spells are typically short We examine explanations from existing literature Institutional setting rules-out cash management and credit limit management as a plausible explanation Co-holding is only slightly higher among households than individuals Evidence suggests limited attention does not explain co-holding

I Positive relationship between level and duration of co-holding I Co-holding is more likely on days when individuals pay attention to

their accounts (measured by logins)

7/56

Results

Our results suggest co-holding may arise due to mental accounting Individuals might co-hold through a desire to separate their financial assets from their short-term consumption (Thaler, 1985) We find that the likelihood of coholding increases with share of non-durable expenditure Likelihood increases with share of spending on gambling and temptation goods (consistent with Bertaut et al, 2009; Gathergood and Weber, 2014) Coholding decreases on payday as overdraft “spending” account is replenished by income receipt Indicates individuals may run short-term “spending” (overdraft) and “savings” (deposits) accounts concurrently

8/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

9/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

10/56

Data

Financial aggregator data provided by Meniga Serves approximately 20% of the Icelandic adult population Meniga scrapes transaction-line level data from financial providers and creates a view of accounts on its platform One-time sign-up allows Meniga to scrape data on an ongoing basis Objective transaction records, avoiding non-response or self-report

I Used in previous studies of individual and household financial decisions

(Olafsson and Pagel 2017, 2018a, 2018b; Carvalho et al., 2019)

Data covers period 1 Sept 2014 - 31 Jan 2017 We analyse data at the individual × day level

11/56

Sample Selection

Steps in sample selection Starting sample of 53,300 individuals Sample restrictions

I At least one deposit account and one overdraft line I Minimum of one monthly income receipt I Minimum level of basic transactions (5 in each of 23/24 months) I Available demographic information (age, sex, postal code)

Resulting sample of 11,551 individuals and 10.2 million individual × day observations

12/56

Measurement

Co-holding calculation Two components: cash deposit balances and overdraft balances

I On average, 13pp margin between credit interest rates paid on balances

and overdraft interest rates charged on balances

I Both are fully liquid, can be drawn-down / paid-down instantly I Overdraft balances incur interest on a daily basis

Co-holding is therefore the minima of on-the-day cash deposit balances and (absolute) value of overdraft balances Calculate co-holding for each individual × day observation

13/56

Measurement

Co-holding calculation Calculate an economically meaningful measure of co-holding Scale co-holding (in currency units) by permanent income Consumption value of co-holding differs across households in line with their permanent income

I For high income households a given monetary amount of co-holding i)

is worth less in consumption, ii) has shorter implied duration

Use average daily expenditure over sample period as proxy for permanent income

I Resulting measure of “consumption-days” of co-holding

Also show results for co-holding in monetary units

14/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

15/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

16/56

Measuring Co-holding

Figure: Overdraft Balances

17/56

Measuring Co-holding

Figure: Cash Deposit Balances

18/56

Measuring Co-holding

Figure: Overdrafts vs. Cash Deposit Balances

19/56

Measuring Co-holding

Table: Co-holding in Consumption-Days

Cash holdings Overdraft holdings >0-10 >10-20 >20-30 >30 Total 2.95 15.65 6.79 4.88 35.47 65.73 >0-10 1.65 1.27 0.19 0.10 0.63 3.83 >10-20 1.23 1.12 0.15 0.07 0.35 2.93 >20-30 1.16 1.04 0.14 0.07 0.28 2.70 >30 11.50 9.90 1.02 0.55 1.83 24.81 Total 18.50 28.98 8.29 5.67 38.56 100.00

20/56

Measuring Co-holding

Summary statistics for co-holding 85% of observations have zero co-holding

I 65.7% have overdraft balance equal to zero I 18.5% have cash deposit balance equal to zero I 2.9% have both balances equal to zero

15% of observations have non-zero coholding. Among which:

I 12.3% co-hold over 30 days of overdraft alongside some cash holdings I But only 1.8% co-hold over 30 days of overdraft and cash balances I 7.5% co-hold more than ISK20,000 / $200 I But only 4% co-hold more than $800 21/56

Measuring Co-holding

Table: Co-holding in Monetary Amounts

Cash Overdraft >0-20,000 >20,000-40,000 >40,000-60,000 >60,000-80,000 >80,000 Total 2.95 8.82 3.76 2.92 2.40 44.87 65.73 >0-20,000 0.60 0.38 0.05 0.03 0.03 0.35 1.43 >20,000-40,000 0.45 0.30 0.04 0.03 0.02 0.23 1.08 >40,000-60,000 0.41 0.31 0.05 0.02 0.03 0.20 1.01 >60,000-80,000 0.39 0.26 0.04 0.02 0.02 0.17 0.90 >80,000 13.70 9.57 1.22 0.79 0.57 4.00 29.84 Total 18.50 19.64 5.15 3.81 3.07 49.83 100.00 22/56

Individual-Level Persistence

Summarize co-holding at the individual level Define “co-holder” by at least 1 consumption-day for at least 1 day of the data period. 60% of individuals never co-hold, of the 40%: 1 The fraction of days on which individuals co-hold is typically very low

I 50% co-hold on fewer than 10% of the days in the sample period I Only 15% co-hold on more than half of the days in the sample period

2 The share of days on which individuals co-hold many days of consumption is typically very low

I On average, co-holders hold in excess of 30 consumption-days of

deposit balances and overdrafts on only 11% of their co-holding days

3 There is an inverse relationship between duration of coholding and number of coholding spells

23/56

Individual-Level Persistence

Figure: Share of Days Co-holding

24/56

Individual-Level Persistence

Table: Share of Co-holding Days

Mean SD p50 p75 p90 p95 Min(3,3) 0.216 0.255 0.109 0.325 0.636 0.808 Min(5,5) 0.196 0.242 0.090 0.283 0.588 0.765 Min(10,10) 0.165 0.220 0.064 0.226 0.492 0.699 Min(15,15) 0.146 0.207 0.051 0.195 0.441 0.637 Min(20,20) 0.134 0.197 0.043 0.176 0.419 0.596 Min(25,25) 0.124 0.190 0.037 0.155 0.394 0.566 Min(30,30) 0.117 0.182 0.036 0.143 0.369 0.514

25/56

Individual-Level Persistence

Figure: Duration of Co-holding Spells

26/56

Individual-Level Persistence

Figure: Co-holding Spell Length vs. Frequency

27/56

Individual-Level Persistence

Table: Duration of Co-holding Spells

Duration of Co-holding (#days) Mean SD P50 P75 P90 P95 #obs Individual x day level Individual level 22.5 29.6 52.5 61.4 9.0 13.0 23.0 27.6 43.0 65.5 85.0 112.2 33,841 3,985

28/56

Financial Cost of Co-holding

Calculate financial cost at the individual level for co-holders (60% of sample are non co-holders and hence incur zero costs) We calculate financial cost by multiplying amount co-held by difference in interest rates e.g. ISK 1,000 co-held on a 13%:1% margin creates a 12% excess interest cost per annum, ISK 120 Average annualised costs among co-holders are approximately ISK 4,700 ($47); median is ISK 1200 ($12). 5% of individuals incur more than ISK 21,000 ($210) of excess interest charges

29/56

Financial Cost of Co-holding

Table: Cost of Co-holding for Co-holders

Mean SD P50 P75 P90 P95 #obs Individual × day: Annualized daily costs 4,702 16,673 1,403 12,793 26,917 3,522,740 Individual level: Average annual costs 4,702 9,813 1,198 4,548 12,518 20,823 3,522,740 30/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

31/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

32/56

Determinants of Co-holding

Consider variety of hypotheses: 1 Liquidity needs 2 Within-household co-ordination 3 Limited attention 4 Mental accounting

33/56

Determinants of Co-holding

Consider variety of hypotheses: 1 Liquidity needs 2 Within-household co-ordination 3 Limited attention 4 Mental accounting

34/56

Liquidity Needs

Two explanations from previous studies: 1 Cash management (Telyukova & Wright, 2008). Hold cash for expenses not payable using credit card

I Deposit account balances and overdraft lines equally liquid for

transaction purposes

2 Credit link risk (Druedahl & Joergensen, 2018; Gorbachev & Luengo-Prado, 2019). Hold out balances due to risk lender chasing down credit line

I In our setting lenders issuing overdraft lines cannot impose reductions

for accounts in good standing

35/56

Determinants of Co-holding

Consider variety of hypotheses: 1 Liquidity needs 2 Within-household co-ordination 3 Limited attention 4 Mental accounting

36/56

Within-household Coordination

Co-holding among households Lack of coordination within household (e.g. due to differing levels of patience, as in Bertaut et al., 2009; or other coordination frictions) might explain co-holding To test this explanation, we define units of analysis as:

I Individuals (main analysis) I Identifiable in the app as couples (couples)

Individual co-holding is a lower-bound on co-holding in the couples sample

37/56

Individual vs. Household Co-holding

Figure: Singles

38/56

Individual vs. Household Co-holding

Figure: Couples

100 200 300 200 400 600

Cash holdings in consumption days Overdraft holdings in consumption days

39/56

Within-household Coordination

Evidence for household-level co-holding 15% co-holding on individual × days rises to 21% for couple × days Co-holding more than 40 days overdraft with some cash:

I 9% individual × days I 13% couple × days

Co-holding more than 40 days of both overdraft and cash:

I 1.1% individual × days I 1.6% couple × days

Duration of co-holding is on average shorter among couples × day sample compared with individuals × day sample

40/56

Within-household Coordination

Table: Duration of Co-holding Spells

Duration of co-holding (days) Mean SD P50 P75 P90 P95 #obs Single individual x day level Single individual level Household x day level Household level 22.6 29.9 16.1 21.4 52.8 61.5 41.2 52.2 9.0 13.0 6.0 10.8 24.0 28.0 16.0 19.3 43.0 67.0 29.0 37.3 85.0 112.0 59.0 63.2 30,425 3,563 7,160 524

41/56

Determinants of Co-holding

Consider variety of hypotheses: 1 Liquidity needs 2 Within-household co-ordination 3 Limited attention 4 Mental accounting

42/56

Limited Attention

Limited attention explanation Co-holding may occur to to fixed costs of adjustment, such as time or psychic costs (Sims 2003, 2006) Under this hypothesis we should observe:

I Negative relationship between amount of coholding and duration of

coholding (increasing cost)

I Lower likelihood of co-holding when individuals pay attention to their

accounts, e.g. login to the platform

We observe the opposite relationships to those predicted by a model

• f limited attention

43/56

Limited Attention

Figure: Level of Co-holding vs. Duration

44/56

Determinants of Co-holding

Consider variety of hypotheses: 1 Liquidity needs 2 Within-household co-ordination 3 Limited attention 4 Mental accounting

45/56

Mental Accounting

Coholding might arise due to individuals organising their separate financial accounts as mental accounts E.g. in model of Prelec and Loewenstein (1998) individuals assign durable and non-durable purchases to separate mental accounts Quispe-Torreblanca et al. (2019) find evidence from credit card revolving balances consistent with this model Individuals might co-hold through a desire to separate their financial assets from their short-term consumption (Thaler, 1985) We explore this using information on transaction types and values

46/56

Mental Accounting

Econometric approach Estimate econometric models to explain the start of a coholding spell, using individual × day observations Key variables of interest are transaction variables

I Payday dummy I Share or durable and non-durable consumption I Dummies for consumption categories e.g. gambling, alcohol

Control variables:

I Demographics: age, gender I Couple dummy variables I Log total income, dummy for receipt of benefits 47/56

Mental Accounting

Results Estimates consistent with main analysis. Coholding is:

I Less likely among couples (households) I Less likely on login-days (attention)

Estimates consistent with explanation based on mental accounting

I Likelihood of coholding increases with non-durable expenditure,

including cash spending

I Stronger effects for gambling and temptation goods (as in Bertaut et

al, 2009; Gathergood and Weber, 2014)

I Coholding decreases on payday as overdraft “spending” account is

replenished by income receipt

48/56

Regression Estimates

Table: OLS Estimates Probability of Co-holding

Probability of Co-hold Period Starting (1) (2) (3) (4) Age 0.0001*** 0.0000*** 0.0001*** 0.0001*** (0.0000) (0.0000) (0.0000) (0.0000) Female

• 0.0001
• 0.0001
• 0.0001
• 0.0001

(0.0001) (0.0001) (0.0001) (0.0001) Linked

• 0.0004***
• 0.0005**
• 0.0003
• 0.0004***

(0.0002) (0.0002) (0.0002) (0.0002) Benefits person 0.0006*** 0.0006*** 0.0005*** 0.0006*** (0.0001) (0.0001) (0.0001) (0.0001) Log total income 0.0008*** 0.0006*** 0.0005*** 0.0005*** (0.0000) (0.0000) (0.0000) (0.0000) Payday

• 0.0043***

(0.0006) Durables 0.0001** (0.0000) Non-durables 0.0006*** (0.0001)

• Nr. current accounts

0.0002** (0.0001)

• Nr. savngs accounts

0.0010*** (0.0000) Log cash spend 0.0003*** (0.0000) Log credit card spend 0.0001*** 49/56

Regression Estimates

Table: OLS Estimates Probability of Co-holding

Probability of Co-hold Period Starting (1) (2) (3) Age 0.0001*** 0.0001*** 0.0000*** (0.0000) (0.0000) (0.0000) Female

• 0.0001
• 0.0001
• 0.0000

(0.0001) (0.0001) (0.0001) Linked

• 0.0004**
• 0.0006***
• 0.0005***

(0.0002) (0.0002) (0.0002) Benefits person 0.0006*** 0.0006*** 0.0006*** (0.0001) (0.0001) (0.0001) Log total income 0.0005*** 0.0005*** 0.0008*** (0.0000) (0.0000) (0.0001) Payday

• 0.0050***

(0.0007) Lottery

• 0.0000
• 0.0011

(0.0006) (0.0007) Gambling 0.0069*** 0.0062*** (0.0017) (0.0019) Temptations 0.0021*** 0.0025*** (0.0006) (0.0007) Alcohol

• 0.0008
• 0.0022***

(0.0007) (0.0008) Logins 0.0015*** 0.0016*** (0.0003) (0.0002) 50/56

Regression Estimates

Table: Individual FE Estimates Probability of Co-holding

Probability of Co-hold Period Starting (1) (2) (3) (4) Log total income 0.0008*** 0.0006*** 0.0005*** 0.0005*** (0.0001) (0.0000) (0.0000) (0.0000) Payday1

• 0.0044***

(0.0010) Durables2 0.0001*** (0.0000) Non-durables2 0.0006*** (0.0001)

• Nr. current accounts

0.0020*** (0.0006)

• Nr. savngs accounts

0.0024*** (0.0003) Log cash spendings 0.0003*** (0.0000) Log credit card spendings 0.0001*** (0.0000) 51/56

Regression Estimates

Table: Individual FE Estimates Probability of Co-holding

Probability of Co-hold Period Starting (1) (2) (3) Log total income 0.0005*** 0.0005*** 0.0008*** (0.0000) (0.0000) (0.0001) Payday1

• 0.0049***

(0.0012) Lottery3 0.0001

• 0.0008

(0.0007) (0.0008) Gambling3 0.0046 0.0034 (0.0028) (0.0030) Temptations3 0.0020*** 0.0023*** (0.0006) (0.0008) Alcohol3

• 0.0006
• 0.0019**

(0.0007) (0.0008) Logins4 0.0019*** 0.0025*** (0.0005) (0.0007) 52/56

Mental Accounting

Discussion Results consistent with mental accounting model in which an “expenditure” account is held separately from a “savings” account The “expenditure” account is used for frequent non-durable expenditure, while the “savings” account is preserved for infrequent, durable expenditure At payday, the expenditure account receives an inflow of monies, hence co-holding decreases This may explain why most coholding is short-lived and occurs in frequent spells

53/56

Overview

1 2 3 4

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding

5 Conclusion

54/56

Overview

1 2 3 4 5

Introduction Data Results I: Measuring Co-holding Results II: Determinants of Co-holding Conclusion

55/56

Conclusion

We examine co-holding behaviour Draw on unique Icelandic data providing daily-level transaction records Institutional setting features highly liquid cash and overdraft accounts, permitting us to examine daily-level co-holding Few individuals co-hold, most co-holding is short-lived and average annualised excess interest charges are modest Transaction-level data analysis suggests co-holding might be explained by individuals holding separate mental accounts for “savings” vs short-term non-durable consumption

56/56