I have a dataset with MTD returns for different securities and I want to compound them over time. Anyone know how I can do this in Tableau?
Ideally I want a timeline showing the returns compounding from month to month. Since I have the ending returns for each month for each security I don't need to calculate this, on the other hand I do not have the dollar value of the securities, so I can not use this in my calculation.
The sample data that can be taken
+--------+--------+
| Month | return |
+--------+--------+
| Jan-19 | 10% |
+--------+--------+
| Feb-19 | 15% |
+--------+--------+
| Mar-19 | 20% |
+--------+--------+
| Apr-19 | 10% |
+--------+--------+
| May-19 | 0% |
+--------+--------+
| Jun-19 | 11% |
+--------+--------+
| Jul-19 | 14% |
+--------+--------+
| Aug-19 | 9% |
+--------+--------+
| Sep-19 | 6% |
+--------+--------+
| Oct-19 | 15% |
+--------+--------+
| Nov-19 | 20% |
+--------+--------+
| Dec-19 | 8% |
+--------+--------+
| Jan-20 | 4% |
+--------+--------+
| Feb-20 | 9% |
+--------+--------+
| Mar-20 | 7% |
+--------+--------+
| Apr-20 | 1% |
+--------+--------+
I want the timeline to show Compounded Monthly Growth Rate (CMGR) as:
August: 10% return
September: 26.5% return (1-(1+0.10)*(1+0.15) = 0.265 OR 26.5%)
October: 51.8% return (1-(1+0.10)*(1+0.15)(1+0.20) = 0.518 OR 51.8%)
Presently I am doing separate calculations for each month, but I am pretty sure there is an easy way in tableau to show this growth rate, by applying some running product (cumulative product) kind of function.
Any help is appreciated!
I have taken this sample data..
+--------+--------+
| Month | return |
+--------+--------+
| Jan-19 | 10% |
+--------+--------+
| Feb-19 | 15% |
+--------+--------+
| Mar-19 | 20% |
+--------+--------+
| Apr-19 | 10% |
+--------+--------+
| May-19 | 0% |
+--------+--------+
| Jun-19 | 11% |
+--------+--------+
| Jul-19 | 14% |
+--------+--------+
| Aug-19 | 9% |
+--------+--------+
| Sep-19 | 6% |
+--------+--------+
| Oct-19 | 15% |
+--------+--------+
| Nov-19 | 20% |
+--------+--------+
| Dec-19 | 8% |
+--------+--------+
| Jan-20 | 4% |
+--------+--------+
| Feb-20 | 9% |
+--------+--------+
| Mar-20 | 7% |
+--------+--------+
| Apr-20 | 1% |
+--------+--------+
Step-1 After importing the data in tableau create a calculated field CMGR with calculation as
EXP(RUNNING_SUM(LN(1+SUM([Return]))))-1
the calculation does running_sum on logarithmic scale giving us running_product like function (which sadly isn't available in tableau at present)
Step-2 Build your view (like the screenshot shows it works as desired)
Or a line chart (if desired)
Related
predicting vehicle utilization :
dears this is my data for the full year (8000 records) ...and like to make prediction for next 3 months
could you please advice which Algorithm should i use and any other advice...(i am beginner)
|Branch |month| date |Util |
|:----- |:---:|:-----------:|-----------:|
|1101 |1 | 2022-01-01 | 43.54 |
|1103 |1 | 2022-01-02 | 74.37 |
|1104 |1 | 2022-01-03 | 0 |
|1126 |2 | 2022-01-04 | 65.83 |
Hellow Team,
I would like to know how to extract the Trial balance from my journal entry data by using Laravel 9 Eloquent:
My Vouchers Table
| id |voucher_date| debit | credit| amount |
|----|------------|-------|-------|-----------|
| 1 | 2021-09-01 | 8 | 2 | 5000.000 |
| 6 | 2021-09-22 | 22 | 17 | 4750.000 |
| 8 | 2021-09-05 | 8 | 3 | 1485.000 |
| 9 | 2021-08-10 | 8 | 6 | 108.000 |
| 10 | 2021-07-07 | 8 | 23 | 98756.000 |
|11 | Etc. | ... |...... |........ |
Accounts table
| id | name | desc | status |
|----|-----------------------------------|-----------------------------------|--------|
| 1 | Assets | Current Assets | 1 |
| 2 | Stockholders equity | Stockholders or Owners equity | 1 |
| 3 | Liability | Liabilities related accounts | 1 |
| 4 | Operating Revenues | Operating Revenues | 1 |
| 5 | Operating Expenses | Operating Expenses | 1 |
| 6 | Non-operating revenues and gains | Non-operating revenues and gains | 1 |
| 7 | Non-operating expenses and losses | Non-operating expenses and losses | 1 |
| 8 | Etc. | More accounts....... | 1 |
My Desired output is like this: (Just an Example)
| Date | Account | Debit | Credit |
|------------|----------------------------------|---------:|----------:|
| 2021-09-01 | Stockholders equity | 0.00 | 5000.00 |
| 2021-09-05 | Liability | 0.00 | 1485.00 |
| 2021-08-10 | Non-operating revenues and gains | 0.00 | 108.00 |
| 2021-07-07 | Land | 0.00 | 98756.00 |
| 2021-02-25 | Land | 21564.00 | 0.00 |
| 2018-07-22 | Land | 3666.00 | 0.00 |
| 2018-05-14 | Non-operating revenues and gains | 0.00 | 489.00 |
| 2018-09-16 | Equipment | 692.00 | 0.00 |
| 2021-04-18 | Non-operating revenues and gains | 4986.00 | 0.00 |
| 2020-04-19 | Land | 4956.00 | 0.00 |
| 2019-03-15 | Buildings Asset | 0.00 | 4988.00 |
| 2019-12-04 | Inventory | 0.00 | 7946.00 |
| 2019-08-25 | Stockholders equity | 0.00 | 19449.00 |
| | | | |
| | Balance |36,990.00 |36,990.00 |
You need to assign a new foreign key to the the voucher table. And then you can simply apply the join to get the desired output. AS you mentioned that you are using debit and credit as foreign key how can they be used to uniquely identify the Vouchers table?
I am new to DAX formulas. I am looking to see how I can do an excel equivalent of a SUMPRODUCT for the following data:
Id | Metric | Weight | Metric times Weight |
1 | 20% | 30% | 6.0%
2 | 10% | 20% | 2.0%
3 | 25% | 20% | 5.0%
4 | 12% | 10% | 1.2%
5 | 15% | 10% | 1.5%
6 | 2% | 10% | 0.2% | |
Net | 84.0% | 100.0% | 84.0% (expected Metric*Weight)
I need 15.9% which is SUMPRODUCT(col1, col2).
The row total needs to be a sumproduct() versus the expected cross product above.
Any hints how I can achieve this?
If the data is on a table called Table1 then the following DAX formula should work:
WeightedAvg := SUMX(Table1, Table1[Metric] * Table1[Weight])
Basically, the function SUMX is iterating over the table Table 1 doing the product of [Metric] * [Weight] then once all the iterations are done, the results are added.
If you have a column with the product (as in your example), then you just need to drag that field to the values area of the PivotTable.
Hope this helps!
Configuration:
Windows 8.1
neo4j-enterprise-2.2.0-M03
cache type: hpc
8Gb RAM
6Gb for JVM Heap (wrapper.java.initmemory=6144 wrapper.java.maxmemory=6144)
5Gb out of 6Gb of JVM Heap for mapped memory (dbms.pagecache.memory=5G)
Model:
Model represents how users navigate through website.
27 522 896 nodes (394Mb)
111 294 796 relationships (3609Mb)
33 906 363 properties (1326Mb)
293 (:Page) nodes
27522603 (:PageView) nodes
0 (:User) nodes (not load yet)
each (:PageView) node connected with (:Page) node
each (:PageView) node connected with next (:PageView) node
each (:PageView) node connected with (:User) node (not yet)
Query
match (:Page {Name:'#########.aspx'})<-[:At]-(:PageView)-[:Next]->(:PageView)-[:At]->(p:Page)
return p.Name,count(*) as count
order by count desc
limit 10;
Profile info:
+------------------------------------------------+
| p.Name | count |
+------------------------------------------------+
| "#####################.aspx" | 5172680 |
| "###############.aspx" | 3846455 |
| "#########.aspx" | 3579022 |
| "###########.aspx" | 3051043 |
| "#############################.aspx" | 1713004 |
| "############.aspx" | 1373928 |
| "############.aspx" | 1338063 |
| "#####.aspx" | 1285447 |
| "###################.aspx" | 884077 |
| "##############.aspx" | 759665 |
+------------------------------------------------+
10 rows
195363 ms
Compiler CYPHER 2.2
Planner COST
Projection(0)
|
+Top
|
+EagerAggregation
|
+Projection(1)
|
+Filter(0)
|
+Expand(All)(0)
|
+Filter(1)
|
+Expand(All)(1)
|
+Filter(2)
|
+Expand(All)(2)
|
+NodeUniqueIndexSeek
+---------------------+---------------+----------+----------+-------------------------------------------+--------------------------------------------------+
| Operator | EstimatedRows | Rows | DbHits | Identifiers | Other |
+---------------------+---------------+----------+----------+-------------------------------------------+--------------------------------------------------+
| Projection(0) | 881 | 10 | 0 | FRESHID105, FRESHID110, count, p.Name | p.Name, count |
| Top | 881 | 10 | 0 | FRESHID105, FRESHID110 | { AUTOINT1}; |
| EagerAggregation | 881 | 173 | 0 | FRESHID105, FRESHID110 | |
| Projection(1) | 776404 | 35941815 | 71883630 | FRESHID105, p | |
| Filter(0) | 776404 | 35941815 | 35941815 | p | (NOT(anon[38] == anon[78]) AND hasLabel(p:Page)) |
| Expand(All)(0) | 776404 | 35941815 | 49287436 | p | ()-[:At]->(p) |
| Filter(1) | 384001 | 13345621 | 13345621 | | hasLabel(anon[67]:PageView) |
| Expand(All)(1) | 384001 | 13345621 | 19478500 | | ()-[:Next]->() |
| Filter(2) | 189923 | 6132879 | 6132879 | | hasLabel(anon[46]:PageView) |
| Expand(All)(2) | 189923 | 6132879 | 6132880 | | ()<-[:At]-() |
| NodeUniqueIndexSeek | 1 | 1 | 1 | | :Page(Name) |
+---------------------+---------------+----------+----------+-------------------------------------------+--------------------------------------------------+
Total database accesses: 202202762
Query without unnecessary labels
match (:Page {Name:'Dashboard.aspx'})<-[:At]-()-[:Next]->()-[:At]->(p)
return p.Name,count(*) as count
order by count desc
limit 10;
Profile info:
+------------------------------------------------+
| p.Name | count |
+------------------------------------------------+
| "#####################.aspx" | 5172680 |
| "###############.aspx" | 3846455 |
| "#########.aspx" | 3579022 |
| "###########.aspx" | 3051043 |
| "#############################.aspx" | 1713004 |
| "############.aspx" | 1373928 |
| "############.aspx" | 1338063 |
| "#####.aspx" | 1285447 |
| "###################.aspx" | 884077 |
| "##############.aspx" | 759665 |
+------------------------------------------------+
10 rows
166751 ms
Compiler CYPHER 2.2
Planner COST
Projection(0)
|
+Top
|
+EagerAggregation
|
+Projection(1)
|
+Filter
|
+Expand(All)(0)
|
+Expand(All)(1)
|
+Expand(All)(2)
|
+NodeUniqueIndexSeek
+---------------------+---------------+----------+----------+-----------------------------------------+---------------------------+
| Operator | EstimatedRows | Rows | DbHits | Identifiers | Other |
+---------------------+---------------+----------+----------+-----------------------------------------+---------------------------+
| Projection(0) | 881 | 10 | 0 | FRESHID82, FRESHID87, count, p.Name | p.Name, count |
| Top | 881 | 10 | 0 | FRESHID82, FRESHID87 | { AUTOINT1}; |
| EagerAggregation | 881 | 173 | 0 | FRESHID82, FRESHID87 | |
| Projection(1) | 776388 | 35941815 | 71883630 | FRESHID82, p | |
| Filter | 776388 | 35941815 | 0 | p | NOT(anon[38] == anon[60]) |
| Expand(All)(0) | 776388 | 35941815 | 49287436 | p | ()-[:At]->(p) |
| Expand(All)(1) | 383997 | 13345621 | 19478500 | | ()-[:Next]->() |
| Expand(All)(2) | 189923 | 6132879 | 6132880 | | ()<-[:At]-() |
| NodeUniqueIndexSeek | 1 | 1 | 1 | | :Page(Name) |
+---------------------+---------------+----------+----------+-----------------------------------------+---------------------------+
Total database accesses: 146782447
Message.log
Question
How can I perform this query much faster? (more RAM, refactor query, distributed cache, use another language/shell/method, ...)
UPD:
Profile info for last query in answer
neo4j-sh (?)$ profile match (:Page {Name:'Dashboard.aspx'})<-[:At]-()-[:Next]->()-[:At]->(p)
with p,count(*) as count
order by count desc
limit 10 return p.Name, count;
+------------------------------------------------+
| p.Name | count |
+------------------------------------------------+
| "OutgoingDocumentsList.aspx" | 5172680 |
| "DocumentPreview.aspx" | 3846455 |
| "Dashboard.aspx" | 3579022 |
| "ActualTasks.aspx" | 3051043 |
| "DocumentFillMissingRequisites.aspx" | 1713004 |
| "EditDocument.aspx" | 1373928 |
| "PaymentsList.aspx" | 1338063 |
| "Login.aspx" | 1285447 |
| "ReportingRequisites.aspx" | 884077 |
| "ContractorInfo.aspx" | 759665 |
+------------------------------------------------+
10 rows
151328 ms
Compiler CYPHER 2.2
Planner COST
Projection
|
+Top
|
+EagerAggregation
|
+Filter
|
+Expand(All)(0)
|
+Expand(All)(1)
|
+Expand(All)(2)
|
+NodeUniqueIndexSeek
+---------------------+---------------+----------+----------+------------------+---------------------------+
| Operator | EstimatedRows | Rows | DbHits | Identifiers | Other |
+---------------------+---------------+----------+----------+------------------+---------------------------+
| Projection | 881 | 10 | 20 | count, p, p.Name | p.Name, count |
| Top | 881 | 10 | 0 | count, p | { AUTOINT1}; count |
| EagerAggregation | 881 | 173 | 0 | count, p | p |
| Filter | 776388 | 35941815 | 0 | p | NOT(anon[38] == anon[60]) |
| Expand(All)(0) | 776388 | 35941815 | 49287436 | p | ()-[:At]->(p) |
| Expand(All)(1) | 383997 | 13345621 | 19478500 | | ()-[:Next]->() |
| Expand(All)(2) | 189923 | 6132879 | 6132880 | | ()<-[:At]-() |
| NodeUniqueIndexSeek | 1 | 1 | 1 | | :Page(Name) |
+---------------------+---------------+----------+----------+------------------+---------------------------+
Total database accesses: 74898837
As I mentioned before, in your other question, if you can write a Java based server extension you can do it pretty easily.
// initialize counters
Map<Node,AtomicInteger> pageCounts = new HashMap<>(300);
for (Node page : graphDb.findNode(Page)) pageCounts.put(page,new AtomicInteger());
// find start page
Label Page = DynamicLabel.label("Page");
Node page = graphDB.findNode(Page,"Name",pageName).iterator().next();
// follow page-view relationships
for (Relationship at : page.getRelationships(At, INCOMING)) {
// follow singular next relationship
Relationship at2 = at.getStartNode().getSingleRelationship(Next,OUTGOING);
if (at2==null) continue;
// follow singular page-view relationship to end-page
Node page2 = at2.getSingleRelationship(At,OUTGOING).getEndNode();
// increment counter
pageCounts.get(page2).incrementAndGet();
}
// sort pages by count descending
List pages = new ArrayList(pageCounts.entrySet())
Collections.sort(pages,new Comparator<Map.Entry<Node,Integer>>() {
public int compare(Map.Entry<Node,Integer> e1, Map.Entry<Node,Integer> e2) {
return - Integer.compare(e1.getValue(),e2.getValue());
}
});
// return top 10
return pages.subList(0,10);
For Cypher I would try something like this:
match (:Page {Name:'#########.aspx'})<-[:At]-(pv:PageView)
WITH distinct pv
MATCH (pv)-[:Next]->(pv2:PageView)
with distinct pv2
match (pv2)-[:At]->(p:Page)
return p.Name,count(*) as count
order by count desc
limit 10;
Update
I wrote a test for it and ran it on my bigger linux machine, the results there are much more sensible: between 1.6s in Java and 5s max in Cypher.
Here is the code and the results: https://gist.github.com/jexp/94f75ddb849f8c41c97c
In Cypher:
-------------------
match (:Page {Name:'Page1'})<-[:At]-()-[:Next]->()-[:At]->(p)
return p.Name,count(*) as count
order by count desc
limit 10;
+-------------------+
| p.Name | count |
+-------------------+
| "Page169" | 975 |
| "Page125" | 959 |
| "Page106" | 955 |
| "Page274" | 951 |
| "Page176" | 947 |
| "Page241" | 944 |
| "Page30" | 942 |
| "Page44" | 938 |
| "Page1" | 938 |
| "Page118" | 938 |
+-------------------+
10 rows
in 3212 ms
[Compiler CYPHER 2.2
Planner COST
+---------------------+---------------+--------+--------+--------------------------+---------------------------+
| Operator | EstimatedRows | Rows | DbHits | Identifiers | Other |
+---------------------+---------------+--------+--------+--------------------------+---------------------------+
| Top | 488 | 10 | 0 | FRESHID71, FRESHID76 | { AUTOINT1}; |
| EagerAggregation | 488 | 300 | 0 | FRESHID71, FRESHID76 | |
| Projection | 238460 | 264828 | 529656 | FRESHID71, p | |
| Filter | 238460 | 264828 | 0 | p | NOT(anon[29] == anon[51]) |
| Expand(All)(0) | 238460 | 264828 | 529656 | p | ()-[:At]->(p) |
| Expand(All)(1) | 238460 | 264828 | 778522 | | ()-[:Next]->() |
| Expand(All)(2) | 476922 | 513694 | 513695 | | ()<-[:At]-() |
| NodeUniqueIndexSeek | 1 | 1 | 1 | | :Page(Name) |
+---------------------+---------------+--------+--------+--------------------------+---------------------------+
Total database accesses: 2351530]
And in Java:
-------------------
Java took 1618 ms
Node[169]=975
Node[125]=959
Node[106]=955
Node[274]=951
Node[176]=947
Node[241]=944
Node[30]=942
Node[1]=938
Node[44]=938
Node[118]=938
Something you can also do to speed up your Cypher query, is to only aggregate on the nodes, and only return the page.Name property for the last 10 rows, much faster.
match (:Page {Name:'Page1'})<-[:At]-()-[:Next]->()-[:At]->(p)
with p,count(*) as count
order by count desc
limit 10 return p.Name, count
So I'm stuck. I am working on a credit system with expirations. Similar to credit card miles but not exactly. By the way I am sorry for the book ahead but I needed to add enough detail to help get the whole picture.
What I need is a system where a user accumulates credits for doing activities. But they can also spend these credits on activities. The credits should expire after 30 days if they are not used. I seem to be stuck on how to accurately calculate this in a batch that will run every night. Any ideas in any language would be greatly appreciated as I seem to be stuck on just one minor detail that I can't get around. Here is an example of the data:
7/1: +5 - user signs up
7/2: +5 - user interacts with system
7/2: -3 - user purchases activity
7/3: +5 - user interacts with system
So at this point the user has received 15 credits and has spent 3. Leaving him with a total of 12 credits. (At least I got basic math down :P)
I should add that currently we are playing with the idea of having two fields: last processed, next processed. So these values at this time assuming it was a new sign up are:
Last Processed Date: 7/1
Next Process Date: 8/1
So now 8/1 comes around. The batch starts and looks at all credits that are older than 30 days. Which at this point is 5.
This is where it starts to get fuzzy.
Then the system should look at all the credits that have been spent in the last 30 days to see if they are using any credits. Because they should only expire if they haven't been used. So there are 3. So I then deduct the user 2 credits because that is the difference of credits earned older than 30 days and what has been spent. So I finish the batch and set the dates accordingly for the next day. Now assuming they haven't spent anymore I start the calculation over of credits earned older than 30, which is 5 and credits spent which again is 3. But I obviously don't want to consider the 3 credits that I considered yesterday. What is a good approach to not include those 3 credits again for consideration.
That is where I am stuck.
We are thinking about writing a debit record for the expired credits so we can track them but having a hard time seeing how I can use it in this calculation.
If you read this far thank you. If you even make a somewhat effort in the answer I will at a minimum give you an up vote for effort.
EDIT:
Ok #Greg mentioned something that I forgot to address. The idea of putting a flag on the credits considered. A valid point but not one that can work because of the following scenario:
Let's say that on a particular day a user spends 10 credits. But the expired credits that the batch is considering only accumulated to 5. Well he should still have 5 more credits left over to not have expired because he spent more than a single expiration. So the flag wouldn't work because we would have skipped those 5 extra credits. Hope that makes sense?
For every user of the system keep an array, that stores information about the amount of credits available to the user for the next 30 consecutive days
For example the data for some user might look like this
8 |
7 | |
6 | | | |
5 | | | | | | | | | | |
4 | | | | | | | | | | | | | | | | |
3 | | | | | | | | | | | | | | | | | | | | | | | |
2 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
1 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
-------------------------------------------------------------
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
^ ^ ^
| \_ |
today tomorrow in 15 days
Every time the user earns some credits, You increase amounts for all days by the number of credits earned. For example if the user earns 2 credits the table changes as follows. It's like rising the whole graph up.
10 |
9 | |
8 | | | |
7 | | | | | | | | | | |
6 | | | | | | | | | | | | | | | | |
5 | | | | | | | | | | | | | | | | | | | | | | | |
4 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
3 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
2 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
1 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
-------------------------------------------------------------
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
^ ^ ^
| \_ |
today tomorrow in 15 days
If The user has x credits today and spends y credits, You decrease the amount of credits available to him to x - y, for every day he has an amount greater than x - y. For days he has no more than x - y, the amount stays the same. It's like cutting the top of the graph off. For example if the user spends 3 credits the graph changes to
7 | | | | | | | | | | |
6 | | | | | | | | | | | | | | | | |
5 | | | | | | | | | | | | | | | | | | | | | | | |
4 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
3 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
2 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
1 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
-------------------------------------------------------------
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
^ ^ ^
| \_ |
today tomorrow in 15 days
Every day You shift the graph to the left to model expiring credits. The user will have the following amounts tomorrow
7 | | | | | | | | | |
6 | | | | | | | | | | | | | | | |
5 | | | | | | | | | | | | | | | | | | | | | | |
4 | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
3 | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
2 | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
1 | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
-------------------------------------------------------------
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
^ ^ ^
| \_ |
today tomorrow in 15 days
I wouldn't consider trying to process the data as you present it. Instead, you should keep track of how many credits the user has, and when they expire. That way you keep track of which credits were used when the purchase is made, instead of trying to work it all out later.
So when the user signs up, they have:
5 credits expiring on 8/1
After interacting with the system the next day:
5 credits expiring on 8/1
5 credits expiring on 8/2
After purchasing something:
2 credits expiring on 8/1
5 credits expiring on 8/2
And so on.
Assuming you run this batch on a daily basis, you can have a table that keeps track of all the credits they earned, and the credits they used (negative credits).
At the beginning of the next month, your job is simply to find out which of the credits earned on the first day were not spent during the month.
The number of credits earned on the first day - the credits they spent all of last month. If the number is positive, they have some credits that need to expired. So simple add a record in the table with a negative credit. This will zero-out the unused credits.
The next day, repeat the process by seeing how many credits they earned on the second day minus the sum of all the credits they earned in the last month, taking into account the record with the negative credits you created the previous day.
How about adding a flag to the expenditures? If the flag is not set, then you can include that expenditure in the batch, if necessary. If you do use the expenditure to offset an expiration, then you set the flag. Next time through, you'll ignore that expenditure because the flag is set.
Use a debit record to record normal expenditures. When the monthly batch job runs, it can calculate the total debits which are less than or equal to the expiring credits. If there are credits to expire, simply insert an appropriate debit record (appropriate == to cancel the excess, in your application). In this way, any 'running total' code which examines only credits and debits will reach the same balance that your batch code intended.
One approach to this problem is to store only the transactions, not the balance. Then you always calculate the balance in real time when needed. Here's the data:
Date : Amount : Expiries
7/1 : +5 : 7/31
7/2 : +5 : 8/1
7/2 : -3 : never
7/3 : +5 : 8/2
The balance at any time is simply the total of all transactions that have not yet expired. No need to run any batch processes.
Regarding Julians reply (that I can't comment to yet), I'm dealing with just the same problem and Julians approach won't work because that would result the account being able to go negative.
If the user didn't use the service for one month, on 8/4 the account balance would be -3 and one activity worth of 5 would bring the balance to 2, not to 5 as it should.