# Trustworthiness at the Border

# Participants
Surajit Ray ([e-mail])
Adhiraj Mandal ([e-mail])
Sabrina Kombrink ([e-mail])
Tony Samuel ([e-mail])
Yutong Bai ([e-mail])
Sophie Abrahams ([e-mail])

# Problem-description

This subgroup will look at modelling trust at the UK boarders, focusing on the relationship between the importer and the government.

As explained in the UK Border Strategy (UK Government, 2020) report, the reducing friction in International Trade (RFIT) project was initiated in March 2019 to understand how Blockchain Distributed Ledger Technology and associated technologies can be used to seamlessly integrate supply chain data with HM Revenue and Customs (HMRC) and the Food Standards Agency’s systems, guaranteeing the timeliness and provenance of critical data and avoiding the need for discrete declarations. This project is a step towards a more operationally efficient supply chain network, as it provides transparency in the data between producers, importers and the UK border.

There is typically a mature relationship between an importer and a producer. However, this is not necessarily the case for the relationship between an importer and the UK government. We therefore plan to model the latter during the study group. This model is expected to incorporate attributes such a calculative trust and cognitive trust etc., as detailed in the paper: The impact of a blockchain platform on trust in established relationships: a case study of wine supply chains, while also considering uncertainties. Other aspects to do with operational resilience, developments in technology and how the model fits with the complex supply chain (where there's a network of trust) could also be considered.

Together with the HMRC border risk assessment, the model developed during the study group could then be used by relevant parties (e.g., the government) to help determine how trustworthy a given importer is, and thus help manage risk at the borders.

# Questions to address
**Aim:** Develop a model with scoring mechanism.

1. How do we measure the variables?
2. What type of model do we wish to fit?
3. Do we have data at hand?
4. Are we supposed to develop a questionnire or develop a scoring technique for a statistical (possibly Bayesian) model?
5. More explanations on the diagram on pg 132. Do we have to expand this diagram?
6. Find existing models that describe calculative trust, cognitive trust, etc.

**Main reference:** [1] https://www.emerald.com/insight/content/doi/10.1108/SCM-05-2021-0227/full/pdf?title=the-impact-of-a-blockchain-platform-on-trust-in-established-relationships-a-case-study-of-wine-supply-chains

# Notes from first day of reading

### Types of trust as defined in Reference [1]:
1. **Calculative trust:** dominates early in relationships. "Data integrity and information asymmetries characterise transactions where cost, benefit and reputation are core drivers"
1. **Cognitive trust:** develops as trust builds progressively. "Combines transactional and relational elements, expressed by expectations and predictions that partners will meet obligations. "
1. **Affective trust:** Occurs as the relationship matures. "Builds on shared values, creating reciprocal durable personal attachments between buyers and sellers with behavioural trust. Reliance on others and disclosure of confidential information are key trusting behaviours."

### Types and properties of trust according to Reference [2]:

**Types of trusts**
1. **Cognitive trust** – generalized expectancy held by an individual that the word of another can be relied on
2. **Calculative trust** – trust grounded in the rational calculation of the costs and benefits of another individual breaking and maintaining an interdependent relationship. Calculative Trust (CT) is modelled in this paper: https://core.ac.uk/download/pdf/80962048.pdf
3. **Authorised Economic Operator (AEO) certification** - highest level of trust from HMRC?

**Properties of trust**

* **Transitivity:** Due to the lack of previous experience, the central role in establishing trust is given to the brand, reputation of enterprises, references and recommendations by the partner with some history of cooperation with the enterprise, which is referred to as the principle of transitive trust (i.e. if A trusts B, and B trusts C, then A can trust C).

## Other notes from Reference [1]:
* Trust between importer/logistics company and UK Government agencies remains at calculative level. This is based on risk, compliance and history.
* “The default position is of limited trust, trust is built up over a number of years to partnerships, shared values, based on integrity and personal reputation” (Imp)
* “The UK Government customs and excise regime is based on approving key operators in the supply chain.” (HMRC)
* “There is a higher level of trust in the information when it is a highly regulated product. Where there is a balance between trust, risk and reputation.” (Imp)
* More information on p139 about how trust is built. Trust is built on past interactions.
* Errors in data (common due to data being manually entered) causes distrust.
* Distrust causes friction at borders.

# Insights from the stakeholder

**Aims:** Develop a model with scoring mechanism. Look at trust attributes, how can we measure some of these attributes (maybe red/amber/green trust)? Look at different trusts, what is good, what is bad? Then, we can model where an operator is today. Build a model that demonstarates the type of interactions between all actors, this would be important for small importers. How can you build up trust for small importers.
- Trustworthiness is improved over time with good experience.
- Someone new to border operations has a very low level of trust.
- Someone with a high level of movements/deliveries may still not have a mature level of trust. Look at transporting companies like UPS, they do many movements per week, but they do not reach the high level of trust as even with such large volumes, they have discrepancy issues - wrong data, wrong commodity code entered in the declaration, that results in rejection at the border.
- Calculative trust is based on very low level. Cognitive trust has more attributes. Trust is built over time.
- Any error (e.g., data entry error) results in mistrust at the border.
- Small steps up, big step down. (Step up 50, but down 70 because you have made an error.)
- How should we increase/decrease trust:
- If an importer decides to import goods themselves, and then struggles with the information required at the border, then they are fundamental not trusted.
- If there is an error in the paper work or the operation, then the goods are stuck in the port. Goods stuck in the port is a worst case scenario - if the goods are food, then they can spoil, or there are customer satisfaction issues.
- Small importers may opt to use a third party which is more trusted, as they are doing more imports / successful completions of declarations.
- The more successful declarations a company completes, the more trusted they are. Once an importer gets over a certain threshold, then the likelihood of goods getting stopped/rejected at the border falls. Goods not examied going through customs if trust is high.
- Gov do not give out their trust metrics. They have declarations. AEO: once an importer has given a certain number of declarations, then they get an AEO 'badge'. An error in a declaration results in mistrust. If we build a model that can demonstrate this interatction, then this would be very important for logistics companies. (AEO status is a certified standard authorisation issued by customs administrations in the European Union (EU))

# Previous work, information to build model on

### Further references
* How does existing work on simplex, linear supply chains (Lewicki et al., 2006; Poppo et al., 2016; Tejpal et al., 2013) look like ? -- Assign someone to read and report back
* Complex multiparty supply chains (Chen et al., 2019) limited work
* Shin and Bianco (2020) use qualitative methods to determine the affordance of trust in blockchain media acceptance, focussing on trust in technology and transactions, not end-to-end supply chains. - report back

### Data of Y/N answer in questionaire, reference [1]

In Table 2 of paper [Impact of a blockchain platform on trust](https://www.emerald.com/insight/content/doi/10.1108/SCM-05-2021-0227/full/pdf?title=the-impact-of-a-blockchain-platform-on-trust-in-established-relationships-a-case-study-of-wine-supply-chains)

* why only Y/N question .. why not 5 point scale?
* Where is this data?
* Has this data been visualised?
- If not what techniques are available ?

From reference [1]:


**Question?**
Circle -- definitely not directly measurable
rectangle - Are they directly measurable

### Some similarity with structural equations model
Paper: [Structural Equation Model of Trust and Partnering Success](https://ascelibrary.org/doi/epdf/10.1061/%28ASCE%290742-597X%282005%2921%3A2%2870%29)


Sample Questionaire:


Another model from [3]

From [4]


### Existing problem in model selection of SEM

1. relationship
2. Variable selection
3. Model goodness of fit (AIC-BIC)
4. Preparing appropriate questions to answer objectively

### Other possible connections

* Network Analysis

### Notes from article: Heping Chen and Chunjie Qi. The Trade Effect of Trust: Evidence from Agricultural Trade between China and Its Partners

https://www.google.co.uk/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&ved=2ahUKEwjn4Pu8oOv2AhW0QkEAHewUAk0QFnoECBoQAQ&url=https%3A%2F%2Fwww.mdpi.com%2F2071-1050%2F14%2F2%2F729%2Fpdf&usg=AOvVaw131BpjdQpcHsslbFrdG0VT

According to the *trade gravity model with trust as additional component* trust significantly influences the trade volume:

\begin{equation} lntrade_{ijt} = \alpha + \beta lntrust_{jt} + \sum_{k=1}^n \gamma_k X^k + \epsilon\end{equation}

where,
* $lntrade$: dependent variable (trade volume)
* $lntrust$: Core explanatory variable (trust level of trading partners)
* $i$ is the exporting country,
* $j$ is the trading partner,
* $t$ is time,
* $X^k$ is the $k$-th dimension control variable (e.g. gross domestic products of trading countries, geographical distance between trading countries),
* $n$ is the number of control variables,
* $\alpha$ is the constant term,
* $\beta$ and $\gamma_k$ are parameters to be estimated, and
* $\epsilon$ is the error term.

### Notes from articles for measuring variables

**Dubey et al. (2019) - [Swift trust and commitment: The missing links for humanitarian supply chain coordination?](https://link.springer.com/article/10.1007/s10479-017-2676-z)**
Their research framework using path analytic model:

Measuring elements by survey - five-point Likert scale from strongly disagree (1) to strongly agree (5):

### Wang and Singh 2010 - [Evidence-Based Trust: A Mathematical Model Geared for Multiagent Systems ](https://https://ezproxy-prd.bodleian.ox.ac.uk:3400/doi/pdf/10.1145%2F1867713.1867715)

Model of trust based on probability distribution of probability of a postive outcome. Begin with uniform probability and update with each interaction.

Trust is affected by 'evidence' and by 'conflict'. Evidence can be postive and negative. Conflict is great when there is an equal amount of negative and positive evidence.

# Links to data

https://www.ons.gov.uk/economy/nationalaccounts/balanceofpayments/methodologies/uktrade
https://data.oecd.org/trade/trade-in-goods-and-services.htm#indicator-chart

# Modelling

## Modelling variables

**Input variables**

| Input Variable | Type | Simulation weight | Comment |
| ------- | -------- | -------- |-------- |
| Time for transaction | `Numeric in days` | -1 | Might not be relevant Poisson
| Time at border (wait time<2 days) | `Binary` | 2 |
| Number /Proportion of success | `Integer>0, Real between 0 and 1` | 2 | Poisson
| Number of unsuccessful transaction over last year | `Real between 0 and 1` | -10 |
| Type of food | `Fixed for now` But can be modeled | Text |
| Whether using Technology | `Binary` | Text |
| text | text | Text |

**Output Variable**

Trustworthiness score
and threshold for Binary Trust

# Different modelling techniques

### 1. Standard Linear Model

### 2. Path Analysis

leading to

### 3. Structural Equations Model

Reference: https://m-clark.github.io/sem/sem.html#r-packages-used-2
#### Initial model and result
i: "Experience-based trust" from x1,x3,x4
s: Trust from all variables/indicators
Model:
Model with weights and result:

#### Another modified version based on theoretical model in the original paper:
x1=Time for transaction Poisson with mean 3
x2=Time at border (wait time<2 days)
x3=Number of success over last 3 years. ->Poisson with mean 40
x4=Number of unsuccessful transaction over last year ->Poisson with mean 5
x5= technology use
x6 = trustworthy AEO importer
x7 = Proper documentation
Cognitive trust (CoT) indicator: x2
Calculative trust (CaT) indicators: x5, x6, x7
Affective trust(AfT) indicators: x1, x3, x4

### 4. Agent based models

#### Intro to Agent Based Models
"An agent-based model (ABM) is a computational model for simulating the actions and interactions of autonomous agents (both individual or collective entities such as organizations or groups) in order to understand the behavior of a system and what governs its outcomes. It combines elements of game theory, complex systems, emergence, computational sociology, multi-agent systems, and evolutionary programming. Monte Carlo methods are used to understand the stochasticity of these models." From Wikipedia https://en.wikipedia.org/wiki/Agent-based_model

Agent-based models are used for a diverse range of applications such as: the spread of disease in a country; the development of friendships in a society; and the movement of cars at a junction. The simulations show the evolution of the systems over time

#### Application to modelling trust

As agent-based modelling is time-dependent, it is a useful tool for modelling the development of trust over time. We begin with aspects of a structural equations type model to model how 'inputs' (in this case called 'attributes') can be used to determine a trust score for an importer. To this framework, we introuduce elements of randomness to observe how the trust score of an importer can change over time. The agent-based model gives a visual output, marking an importer as red, orange or green based on their trust score.

To model trust, we use the framework of a dynamic network. In our initial example the network consists of two importers, each marked with a node on the network. Each importer has three 'attributes': number of previous sucessful imports; number of previous unsucessful imports; and delay on their most recent import. Each of these three attributes is represented by a node on the network. The weight of an edge on the network between an attribute and an importer is the value of that attribute for that importer. Number of previous unsucessful and sucessful imports are positive integers. Delay on most recent import is given the value 1 for no delay, 0 for a moderate delay and -1 for a severe delay.

Our model is time-dependent and the trust value changes with every attempted import. At each timestep, one importer attempts an import:
* the processing time at the border depends on the current trust score of the importer
* if the current trust score is high, then with high probability there are no delays, or moderate delays
* if the current trust score is low, then with high probability there is a delay, or a moderate delay
* a delay decreases the trust score by 1
* a moderate delay leaves the trust score unchanged
* no delays increases the trust score by 1
* similarly, the sucess of the import depends on the current trust score of the importer
* if the current trust score is high, then with high probability the import is sucessful
* if the current trust score is low, then with high probability the import is unsucessful
* if the import is successful, the trust score increases by 2
* if the import is unsuccessful, the trust decreases by 10

This network can easily be extended to include more importers and more attributes. Interactions between importers could be taken into account, for example, modelling the case where a small importer attemps to import via a large importer. The benefit of this model is that mutual dependencies can be implemented (e.g. the trust score influences the processing time at the border, which in turn influences the trust).

Our model is implemented in **NetLogo**, where the time-dependent simulation can be run.

Alternatively, an implementation in R is possible. **R Packages** that can be used:

lavaan
semTools
semPlot

**Outlook:**
Consider the scenario where there's AEO certificates, where data from other countries is used (transitivity), where interactions between small and large importers are modeled more realistically.

**References:** An agent-based model of supply chains with dynamic structures – https://www.sciencedirect.com/science/article/pii/S0307904X12006592

# References

1. Mike Brookbanks and Glenn Parry. "The impact of a blockchain platform on trust in established relationships: a case study of wine supply chains". *Supply Chain Management: An International Journal* 27/7 (2022) pp.128–146. https://www.emerald.com/insight/content/doi/10.1108/SCM-05-2021-0227/full/pdf?title=the-impact-of-a-blockchain-platform-on-trust-in-established-relationships-a-case-study-of-wine-supply-chains

2. AKRUT, H. (2015), "A process perspective on trust in buyer–supplier relationships. “Calculus”: An intrinsic component of trust evolution", European Business Review, Vol. 27 No. 1, pp. 17-33. https://doi.org/10.1108/EBR-01-2014-0006 or
https://www.emerald.com/insight/content/doi/10.1108/EBR-01-2014-0006/full/pdf?title=a-process-perspective-on-trust-in-buyersupplier-relationships-calculus-an-intrinsic-component-of-trust-evolution

2. Ryciuk, U., & Nazarko, J. (2020). MODEL OF TRUST-BASED COOPERATIVE RELATIONSHIPS IN A SUPPLY CHAIN. Journal of Business Economics and Management, 21, 1225-1247.

3. Joo, Jaehun, and Yuming Han. 2021. "An Evidence of Distributed Trust in Blockchain-Based Sustainable Food Supply Chain" Sustainability 13, no. 19: 10980. https://doi.org/10.3390/su131910980

4. When Can You Trust ‘Trust'? Calculative Trust, Relational Trust,
and Supplier Performance https://core.ac.uk/download/pdf/80962048.pdf

5. WANG & SINGH (2010) North Carolina State UniEvidence-Based Trust: A Mathematical Model Geared for Multiagent Systems. ACM Transactions on Autonomous and Adaptive Systems Vol. 5, No. 4, https://ezproxy-prd.bodleian.ox.ac.uk:3400/doi/pdf/10.1145%2F1867713.1867715

---

@article{Ryciuk2020MODELOT,
title={MODEL OF TRUST-BASED COOPERATIVE RELATIONSHIPS IN A SUPPLY CHAIN},
author={Urszula Ryciuk and Joanicjusz Nazarko},
journal={Journal of Business Economics and Management},
year={2020},
volume={21},
pages={1225-1247}
}

@article{su131910980,
article-number = {10980},
author = {Joo, Jaehun and Han, Yuming},
doi = {10.3390/su131910980},
issn = {2071-1050},
journal = {Sustainability},
number = {19},
title = {An Evidence of Distributed Trust in Blockchain-Based Sustainable Food Supply Chain},
url = {https://www.mdpi.com/2071-1050/13/19/10980},
volume = {13},
year = {2021},
Bdsk-Url-1 = {https://www.mdpi.com/2071-1050/13/19/10980},
Bdsk-Url-2 = {https://doi.org/10.3390/su131910980}}