September 29, 2026 · 3 min read
NAV per unit in Fineract: how an investor comes in and out without diluting anyone
Fineract ships no investment fund product, but NAV-per-unit mechanics sit on top of savings accounts. How it works, with the worked example you should verify before writing any code.
A fund manager who wants to run a fund on Apache Fineract finds out quickly that there is no product to buy: m_fund exists, but it is a catalogue of loan funding sources. No unit price, no participants, no subscriptions. The fund has to be built.
The good news is that the parts are there. Each investor comes in through a Fineract savings account linked to the fund, their deposits and withdrawals are ordinary savings transactions, and units are derived from them. What you have to get right is the price each movement converts at, because that single decision is what makes the fund share out correctly or not.
The unit price and the rule that holds it up
At every close, the unit price is fund value divided by units outstanding. That is all.
The rule that holds everything else up is that a subscription converts at the previous close’s price, not at the current day’s. It feels backwards — today’s price seems fairer — and it is exactly the other way round: using today’s price means computing fund value with the money that just arrived already inside it, then issuing units at that inflated price. The investor coming in takes a share of the return the ones already there produced.
The example to verify
A fund with an initial unit price of 100 and two investors, A and B.
Day 1, launch. A subscribes 60,000 and B subscribes 40,000. That issues 100,000 / 100 = 1,000 units: A holds 600, B holds 400. Fund value is 100,000 and the price stays at 100.
Day 2, the fund earns. It places a portfolio and collects interest. Value rises to 102,000 with no participant activity. Units are still 1,000, so the price moves to 102,000 / 1,000 = 102. The unit returned 2%, and that return belongs to whoever was inside.
Day 3, one in and one out. A subscribes 20,400 and B redeems 10,200, on the same day. Both convert at the previous close’s price, 102: A receives exactly 200 units, B redeems 100. Units outstanding land at 1,100 and fund value at 112,200.
price = 112,200 / 1,100 = 102
The price did not move. That is the point: the capital coming in and the units issued grow in the same proportion, so A does not dilute B on the way in and B does not hurt A on the way out. Positions settle at 800 and 300 units, 72.73% and 27.27% of the fund.
If the price moves on day 3 in your implementation, you have a distribution problem, not a rounding one. Stop there before building anything on top.
What has to be stored
Three things, none of them optional.
The fund’s daily close: date, value, units outstanding, resulting price and the previous close’s price. Storing the previous price on the row itself looks redundant and is not: it is what lets you reconstruct the price that day’s movements converted at without going looking for it, and what makes the calculation auditable months later.
Each transaction’s units, linked to the savings transaction that produced them. It is the only way to answer what was allocated to which investor and when.
The sign of the movement. If units are always stored positive and the sign is inferred from the savings transaction type — deposits add, withdrawals subtract — every query has to replicate that logic. It is a minor detail on day one and a source of disagreement between reports six months later.
A control figure, from day one
At every close, price times units outstanding has to equal fund value. It is an identity, not an approximation: if it does not match, something was computed wrong.
Store it as a column on the close rather than recomputing it on read. It costs one field and gives you an alarm that goes off by itself, at the moment the error happens rather than when someone reviews a statement.
Before building on top
Verify day 3 with your own numbers, in a spreadsheet, before writing the first line. A simultaneous subscription and redemption cannot move the unit price. If they do, the rest of the system inherits that error at every subsequent close, and it surfaces when a participant asks why their return does not match the fund’s.
With the mechanics settled, the hard part is left, and it is not mathematical but architectural: where fund value comes from, where each investor’s position lives, and whether history can be rewritten. That is what decides whether the model survives an audit.