Create a Credit Risk Dashboard in SAS Viya: Part 2

[#1]

In Part 1, we prepped the data for a credit risk dashboard. With the data prepared and promoted, you can now switch to SAS Visual Analytics to create the dashboard.

Step 1: Launch SAS Visual Analytics

Open SAS Visual Analytics via the SAS Viya landing page or hamburger menu, labeled Explore and Visualize. Once open, you will see the Visual Analytics start page or an interface with an option to create a new report. 

Step 2: Create a New Report and Add Data

• Start a New Report: Click the New Report button on the top right. A blank report canvas will appear. You’ll see an empty canvas in the center, a pane on the left for Data and Objects, and on the right an Options/Properties pane for the selected item.
• Add Data Source: In the left Data pane, click Add Data. In the data selection dialog, search for the table you loaded. You should find account_monthly_features. If you do not see it in the data source selector, wait about 30-60 seconds, then try again as it may take a moment for the data crawler to pick up the promoted table from SAS code. Once you see it, select it and click Add. The dataset is now loaded into the report.
• Verify Fields: After loading, the Data pane will list all fields in account_monthly_features, categorized into Measures and Categories. Ensure key variables are present. If any field’s role is misassigned (say Customer_ID might appear as a measure if numeric), you can change its classification by right-clicking the data item and changing its role to Category so it is not aggregated.

Quick Data Check: It’s good to do a quick sanity check in Visual Analytics as well:

• Drag a List Table object from the Objects pane onto the canvas and enable the Detail Data option from the Options pane so that it is not automatically aggregated.
• Add a few columns to it by dragging fields from Data onto the table. You should see rows of data populated.
• Scan through a few records to make sure values look as expected.
• This step ensures the data came through properly. You can remove the list table afterwards as it won’t be part of the final dashboard.

Step 3: Design the Dashboard Layout and Key Metrics

Now, build visuals for the key credit risk metrics on the canvas. We’ll create a set of KPI cards for high-level indicators, a time-series line chart, a distribution chart, and add interactive filters.

A. KPI Metrics Summary

It’s often useful to start the dashboard with a summary of current portfolio risk metrics (like a scoreboard). We can use Key Value objects in Visual Analytics to display single summary values.

• Drag a Key Value object onto the top of the canvas. In its Roles pane, assign a Measure to display. For example, to show average credit score, assign the Credit_Score measure and set the aggregation to Average. 
• Similarly, add another Key Value for Delinquency Rate. If you did not create a Delinquency Rate variable, you can create derive it as a calculated item in Visual Analytics. Use the Delinquent_Flag measure with aggregation Average. Because Delinquent_Flag is 1 for delinquent accounts and 0 for current, its average = proportion of accounts delinquent. Format this as a percentage in the Options pane. Label it “Delinquency Rate".
• Add a Key Value for Expected Loss if you have this available. Assign the Expected_Loss measure with Sum as aggregation. Format as currency with no decimals.
• Optionally, consider adding other KPIs, like Total Accounts or Average Utilization.

Arrange these KPI cards in a single row or a grid at the top of the dashboard for a quick overview. For clarity, you might consider adding an appropriate title to each. At this point, you have a top-line snapshot: which updates based on the data context and will respond to filters we add.

B. Credit Score Trend (Time Series Plot)

One of the requirements is to show average credit score over time. We’ll use a Time Series Plot for this.

• Drag a Time Series Plot object onto the canvas, below the KPI cards. In Roles, set the Time axis (X-axis) to the time variable – Month. Set the Measure (Y-axis) to Credit_Score with an Average aggregation. If you needed a median credit score but have already used it as an average in a key value, you can duplicate the variable and change its aggregation.
• The time series plot will now plot the average credit score by month over the period of data. Each point is the portfolio’s average credit score in that month, and the line connects them to show the trend.
• Customize: give the chart a title like “Average Credit Score Over Time”. If the time axis is too dense, use a hierarchy or select an appropriate interval. In Options, enable markers on data points and perhaps a reference line for a target score if applicable.
• Interpretation check: with this chart, an analyst can see if the average score is trending downward (indicating a possibly riskier portfolio) or upward (improving credit profiles) or seasonally fluctuating.

C. Delinquency Trend

Another key risk indicator is the delinquency rate over time. We can either add another line chart or combine it with the score trend. To keep things simple and clear, we will make a separate line chart for delinquency % by time.

• Drag another Line Chart next to or below the credit score chart. Assign Time axis axis = Month (same time axis). For Measure, assign the Delinquent_Flag measure with aggregation Average. 
• Title this chart “Delinquency Rate Trend”. Format the Y-axis as percentage. The chart will show, for each month, what fraction of accounts were delinquent. You should see a line perhaps hovering in a low percentage, maybe with spikes if there were economic stress periods.
• This visualization allows tracking if delinquencies are rising over time. It complements the credit score chart, since typically if scores drop, delinquencies rise.
• If you prefer, you could overlay this on the same chart as credit score using a dual-axis time series chart for easy comparison. 
• Optional: If you have another time-based measure like Expected Loss over time, you could also plot that in a similar fashion. That would require expected loss by month. Since our data is at month level, we could plot the sum of Expected_Loss on Y and Month on X. That might mimic delinquency trend if PD estimation ties to delinquency, but it’s a useful metric to see if the anticipated loss is growing or shrinking with the portfolio. If you want, add a third line chart for “Expected Loss Trend” with the sum of Expected_Loss by Month, or incorporate it into the delinquency chart via another approach.

D. Credit Score Distribution

We also want to visualize the distribution of credit scores among customers, as an indicator of portfolio risk profile. A Histogram or Box Plot is suitable. Let’s use a histogram:

• Drag a Histogram object onto the canvas. In Roles, assign the Measure as Credit_Score. Visual Analytics will automatically bin the credit score range into system-calculated intervals and show the frequency in each bin. You may want to specify the number of bins yourself using the histogram options to try and get it close to a common credit score range (e.g. bad, acceptable, good, excellent, etc.). Another other option would be to create groupings of credit scores as a new category and assign that to a Bar Chart for more precise control over exact cutoff points.
• Alternatively, a Box Plot could summarize the score distribution (min, median, max, quartiles) succinctly. You could use a box plot with Credit_Score as the measure and perhaps no category or a single category to get an overall box. But a histogram is more informative to see multi-modal distributions or cutoff points.

E. (Optional) Key Risk Indicator by Segment

If your data has categorical segments (e.g., Region, or an internally defined Risk Segment), you might want to include a visual to compare risk metrics by those categories. For example, a Bar Chart of average credit score by Region, or delinquency rate by Segment, could be insightful. Or a bar of Count of accounts by Delinquency_Status category (current vs 30-day vs 60-day, which shows how many are in each bucket). These weren’t explicitly requested, but consider what might be useful: e.g., “Accounts by Delinquency Status” – a bar where categories are Current, 30+ Days Late, 60+ Days, 90+ Days, with count of accounts in each. This gives a quick overview of how many accounts are late and how severely. If you add such a chart, use Delinquency_Status as the category and distinct count of Customer_ID as measure (or just frequency if one record per customer in the latest period). Another example: “Average Utilization by Score Band” – a bar chart where category is score band (we’d need to create a calculated category for score ranges), and measure is avg utilization. This could show that lower-score customers tend to have higher utilization, for instance. These kinds of visuals can enrich the story.

For brevity, we’ll assume the primary visuals are the ones we did: KPIs, score trend, delinquency trend, score distribution.

Step 4: Add Interactive Filters

To make the dashboard interactive and allow drill-down analysis, add filters for key dimensions such as time and customer segments.

• Time Filter: If the dashboard should focus on a particular time window, you can add a filter control for Month or Year. For instance, add a Slider or Drop-down List control. Assign the Measure role to Month or Year. This will allow the user to select a single month or a range of months if using a slider with a range. For example, a slider can act as a timeline to subset the visuals to a specific period.
• Segment/Group Filter: Add a Drop-down control for categorical variables like Region or Segment. Assign Region to it; it will list all regions. The user can select one region to filter the entire report to just customers in that region. If your bank had risk segments, you could filter by that as well.
• Link Controls to Charts: VA’s controls can filter the data at either the report level or just specific objects. The simplest approach is to enable automatic actions so every visual on the page is filtered by the selection. You can do this by selecting the control and in Options --> Actions, then enabling Automatic actions on all objects. Alternatively, for more granularity, you can set up individual Filter or Brush (highlighting) interactions.
• Verify Filtering: Try out the filters: select a particular year or quarter – the KPI values, line charts, and histogram should all update to reflect only data from that period. Similarly, filter by a segment like Region = "West" – now the visuals show metrics for just the Western region’s customers. This interactivity allows risk officers to hone in on specific slices of the portfolio easily.
• Other Controls: If needed, include a Category filter for Delinquency_Status (to see metrics for only delinquent accounts vs all) or Loyalty_Tier if relevant, etc. But don’t overwhelm the dashboard; pick a couple of important filters.
• Other Ways to Filter: Instead of adding controls to the report body, you can add them to the page prompt or report prompt sections at the top of the screen. Expand the section by clicking the down arrow. Controls in page prompts will filter all objects on the page, while controls in report prompts will filter all objects on all pages in the report. You cannot choose which objects are filtered in these sections. You also have the option to change whether it's on the top, bottom, left, or right side of your page or report by selecting the page tab or report title, going to the Options pane and expanding the Report controls section. Note that when adding controls to these sectoins, you must manually enable filtering from one control to another in order to ensure users do not have invalid filtering combinations.

Step 5: Refine and Customize the Dashboard

At this stage, all essential components are in place. Now polish the dashboard for ease of interpretation:

• Layout: Arrange visuals logically. Often, KPIs on top, trends below them, and distribution or breakdown charts to the side. Use containers or sections if supported to group elements. Ensure the charts are sized so that labels and values are readable.
• Titles & Labels: Give each chart a clear title. For axes labels, ensure they are descriptive. If any axis labels or legend entries are not self-explanatory (like "Delinquent_Flag Average"), rename them to plain language. In Visual Analytics, you are not bound to the same variable name rules as in SAS programming, and it is good practice to make data item labels descriptive and clear. This does not change the underlying variable name: only how it is displayed and referenced in Visual Analytics.
• Colors & Display Rules: Apply Display Rules (conditional formatting) to highlight risk thresholds. For example, on the Delinquency Rate KPI, set a rule: if value > 5% then color it red, if < 2% then green. On the credit score trend line, you could color segments of the line red if the average score falls below, say, 660 (subprime threshold). Or simply ensure the line charts use a color scheme that aligns with good/bad (green for credit score trend since up is good, and orange/red for delinquency since up is bad).
• Add Descriptive Text: You might add a text box or an Insights blurb or pop-up window on the dashboard describing what is being shown or any definitions (e.g., “Delinquency Rate = % of customers >30 days past due. Expected Loss is estimated assuming LGD=80%”). This helps business users understand the metrics.
• Test Interactions: Test by clicking on parts of charts if you set up any link actions. For instance, you can allow clicking a bar in the bar chart to act as a filter that highlights or filters other charts to those customers.
• Finalize: Once satisfied, save the report (give it a name like “Credit Risk Dashboard”). You can share it by saving in a shared folder or exporting it.

Now you have a complete Credit Risk Dashboard built in SAS Visual Analytics.

Using the Dashboard for Risk Monitoring and Decisions

With the dashboard components in place, here’s how an analyst can interpret them to drive decisions:

• Portfolio Credit Score Trend: The line chart “Avg Credit Score Over Time” shows the overall credit quality trend. For example, if you see the average score has dropped from 720 to 700 over 6 months, that indicates a deterioration in customer creditworthiness. The risk team should investigate underlying causes – perhaps economic downturn in certain regions or a change in customer mix. A downward trend might prompt tightening credit policies (e.g. reduce credit limits or adjust underwriting criteria) before delinquencies rise. An upward trend would be reassuring, possibly allowing more aggressive marketing or higher limits confidently.
• Delinquency Rate Trend: The delinquency line reveals what fraction of accounts are behind on payments over time. Suppose it spikes in certain months (say it jumps every December and then recovers) – that could be seasonal (holiday spending leading to higher delinquencies) and might normalize. But a steady upward drift in delinquency rate is a warning sign of increasing credit risk in the portfolio. For instance, an increase from 3% to 5% delinquent accounts over a year is significant. The bank might respond by enhancing collection efforts or reviewing accounts that became delinquent to identify common risk factors. If a particular cohort (filterable by the dashboard) is driving the spike (e.g., Region = “East” shows 8% delinquency vs others at 4%), strategies can be region-specific.
• Key Risk Indicator KPIs: The KPI cards at the top give an at-a-glance health check each time the dashboard is opened. For example: Delinquency Rate = 4.5% (maybe set to display red if above 4% via a rule – immediately catching attention). Average Credit Score = 710 (if your internal target is to keep it above 700, being just at 710 might be okay, but if it falls below 700 that card could turn yellow as caution). Expected Loss = $1.2M (if this number has been growing, compare with previous snapshots – you might include a small sparkline in the KPI if Visual Analytics allows, or just watch trends). These high-level numbers support quick decision-making: e.g., if expected losses are climbing quarter over quarter, the finance team may need to allocate more reserves for credit losses (impacting financial statements). If delinquency is low and average score high, perhaps the bank can consider lending growth or lower interest rates competitively, since risk is well-controlled.
• Credit Score Distribution: The histogram shows risk segmentation of customers by credit score. If you observe, for example, a large chunk of customers in the 600-650 bin (which is a borderline segment), the risk team might focus on that segment for preventive action (maybe offer financial counseling, or monitor their behavior closely). If most customers are in 700-800, the portfolio is prime-heavy and relatively safer, but maybe lacking in subprime outreach (if business goal is to also include more subprime for a higher yield, with controlled risk). The distribution may also help in Expected Loss estimation – e.g., the long tail of low-score customers contributes disproportionately to expected loss. It could guide strategies like tightening those accounts or requiring cosigners, etc.
• Filters for Deeper Insight: Using the interactive filters, the analyst can answer specific questions:
• Time filter: Focus on the latest quarter to see current stats (the KPIs and charts will update to just that quarter’s data). This could show, for instance, that Q1 2025 delinquency rate is 5.2% (higher than historical average of ~4%), confirming an uptick recently. Or filter to the 2020 period during a recession to see how bad it got (maybe delinquency hit 8% then). Such insights help in benchmarking and preparing for future scenarios.
• Segment filter: Suppose you select Region = “North”. Immediately, the KPIs and charts update: maybe average score in North is 680 (lower than overall 710), and delinquency 6% (higher than overall 4%). The score trend might show a sharper decline too. This tells you the Northern region is riskier – perhaps an economic issue in that region (like job losses in an industry). Management could respond by reviewing the North region’s credit policies or providing additional support/resources there. Conversely, filtering to “South” might show average score 730 and delinquency 3%, indicating a very healthy sub-portfolio.
• If you had a filter by product type (though here it’s all credit cards presumably), you could see if certain card products have different risk profiles (e.g., premium cards vs standard cards).
• If you implemented a filter by risk band (like choose score < 650 group vs > 750 group), you could then see within the low-score group what the delinquency trend is (likely much higher). This essentially lets you simulate cohort analysis and is useful for targeting interventions.
• Expected Loss Metric: While the expected loss might be shown as a total or average in the KPIs, you can glean insights by observing it in combination with other charts. For instance, if the expected loss is high and rising, check if it correlates with the delinquency trend (expected loss likely incorporates both PD and exposure; rising balances or increasing PDs due to score drops will raise it). If your bank uses the expected loss to set aside loan loss reserves (like CECL – current expected credit loss accounting), a rising trend would lead to higher provisions. The dashboard can thus inform finance how risk trends are translating into dollar impact. If you want to get granular, filtering the dashboard by, say, low score segment will show their portion of expected loss – possibly most of it – highlighting where the risk cost is coming from. This could inform decisions like whether to tighten credit lines for that segment to reduce exposure.

The dashboard provides a comprehensive, real-time view of credit card portfolio risk. Risk managers and analysts can monitor these visuals regularly to catch early warning signs: e.g., “Our average credit score is dipping and delinquency creeping up over the last 3 months in the Midwest region” – armed with this knowledge, they can decide to adjust score cut-offs for new cards in that region, increase collection efforts, or adjust pricing to compensate for higher risk. Conversely, positive trends (improving scores, falling delinquency) could mean it’s safe to expand lending or reduce interest rates to gain market share. The expected loss view ties it all to financial impact, ensuring that decision-makers balance growth and risk appropriately.

Finally, this dashboard is not static – as new data comes in and is processed by your SAS program, (monthly updates of credit scores and delinquencies), the visuals will update, providing an ongoing risk monitoring tool. SAS Visual Analytics makes it easy to refresh the data (if the CAS table is updated or replaced with new data, the report reflects the changes). You can even enable real-time updates to your report with the “Periodically reload report data” option. Over time, the bank can enhance this dashboard, for example by integrating scorecard model outputs or adding drill-down links to customer lists (e.g., click on the “<600” score bar to see a list of those customers for follow-up). But even in its current form, the Credit Risk Dashboard supports proactive credit risk management, letting the bank react quickly to changes and make data-driven decisions to manage the credit card portfolio’s health.

[touwen_k]

@DanielT_SAS I like your fist part a lot, this one would be easier to follow if you have made printscreens of UI. 

[DanielT_SAS]

@touwen_k Good suggestion. We're working on more of these use cases this year, and we appreciate any feedback to make them better.