In short: A replenishment engine solves for quantities and timing in units and holds unit cost as a static field, so it has nothing to contribute to whether you should buy a quarter early under a moving exchange rate. The decision trades an expected currency movement against carrying cost, funding cost, obsolescence risk and the warehouse space the early stock occupies, and all four are measurable from data you already hold. Buying early usually fails because the expected saving gets compared against nothing, when the right comparison is the funding and carrying cost over the extra months held. Using the forward curve rather than a house view settles most of the argument, since the forward already prices the interest rate differential that a one way currency view is implicitly betting against.
You buy from a supplier who invoices in euros, you sell in a currency that lost a third of its value against the euro last year, and the buying team has the same argument every month about whether to bring in the next two quarters now. One side says the currency only travels in one direction and every month of delay costs money. The other side says the warehouse is full and the overdraft is expensive. Both of them are right about something, and the replenishment system has nothing to contribute to either position, because the only question it answers is when stock runs low.
What the replenishment engine can and cannot answer
The engine solves for quantities and timing in units. A reorder point, an order-up-to level, an economic order quantity. Unit cost is a field on the item master, and in most systems it is a standard cost revalued periodically rather than a price that varies with the month the order is placed.
That structure means the objective the engine minimises contains no term for the rate at which the payable gets struck. You can feed a currency forecast into the cost side and get a revalued plan, which changes what the numbers say the inventory is worth. It does not change the recommended order date, because the order date is derived from consumption and lead time. The engine will recommend the same week regardless of what the rate is doing, and it is right to, since nobody gave it a currency objective.
The buying question has a different shape. Given a quantity that has to be available by a particular month, and a window of months in which the order could be placed, which placement month minimises total landed cost in home currency. That is a small optimisation over a handful of candidate dates, and it needs inputs the replenishment engine does not hold.
The components that trade off
Write the total cost of ordering in month t for a requirement in month M as a sum. The sum has a small number of terms and each one is available inside the business.
The currency. The rate that matters is the one at which payment settles, not the one on the day the purchase order is raised, unless you pay at order or hedge at order. For a payable in a hard currency settled ninety days after shipment, the relevant rate lands four or five months after the decision. Getting this wrong is common and it flatters early buying, because it credits the early order with today's rate when the cash actually leaves much later.
Capital cost between payment and sale. Stock bought early is paid for early and then sits. In a high interest rate economy this term dominates everything else on the list. A local borrowing rate of forty percent is roughly three and a third percent a month on the value of the goods, which is a large number to have to overcome.
Warehousing. The marginal cost of the extra space, which is often zero right up until it is very much not zero. Own space has slack to a point, and past that point you are into third party storage at a rate per pallet per week plus the double handling to get stock in and out of it. This cost curve is a step, so it is worth knowing where your step sits before running anything.
Shelf life and obsolescence where they apply. Early arrival consumes remaining life before the stock has done any work, and in categories with a customer-mandated minimum remaining life at delivery, buying early can cost the entire value of the goods rather than a carrying percentage. What to do with the stock once it is short-dated is a disposition question covered elsewhere.
Everything that moves with the calendar rather than the market. Freight rates ahead of a peak season, supplier price list effective dates, factory shutdown weeks, and any announced change to duty or import licensing. These terms are frequently larger than the currency term and they are knowable in advance, which makes them the first thing to check rather than the last.
The case for buying early, and why it usually fails
The intuitive argument is that the local currency depreciates steadily, so every month of delay raises the home currency cost of the same goods. Plotted over three years that argument looks unanswerable.
It usually fails because the depreciation is already priced into the cost of the money you would use to buy early. Covered interest parity says the forward discount on a currency is set by the interest rate differential between the two currencies, and arbitrage holds that relationship closely enough for a purchasing decision wherever both currencies trade freely. If local rates are around forty percent and the hard currency is around four, the forward market is already pricing something like thirty-six percent of annual depreciation. Nobody forecast that figure, it falls out of the arbitrage.
Now run the trade. Buying six weeks early to avoid roughly four percent of expected depreciation costs six weeks of local funding on the full value of the goods, which at forty percent is about four and a half percent, before any storage step. The trade loses, and it loses by construction, because the interest differential that makes the currency depreciate is the same differential you pay to fund the early position.
Early buying wins in specific circumstances, and those are the ones worth going looking for.
You are already funded in the hard currency. If the business holds hard currency deposits or borrows in the supplier's currency, the funding cost is the hard currency rate rather than the local one and the arithmetic reverses completely.
There is an administrative event rather than a market move. An announced tariff change with an effective date, a supplier price list revision, a change in import licensing, a restriction on access to foreign exchange. These are calendar facts, none of them are priced into a forward, and they are the most reliable source of a genuine timing gain.
The supplier's price is fixed in local currency for a window. Buying inside that window is buying at yesterday's rate with no currency exposure at all, so the only cost is carrying.
The currency is managed rather than floating and the peg is under pressure. A step devaluation is a different distribution from a drift, and the expected cost is dominated by a small probability of a large move. This is the one case where a house view is defensible, and it should be written as a probability and a size rather than as a direction.
Use the forward curve rather than a house view
Where a deliverable forward market exists, the forward rate is the right input, for a reason that has little to do with forecasting accuracy.
Meese and Rogoff (1983), in the Journal of International Economics, showed that structural exchange rate models failed to beat a simple random walk out of sample at short horizons, and four decades of re-examination have not overturned that in a way that helps a buyer. Fama (1984), in the Journal of Monetary Economics, showed that forward rates are biased predictors of future spot rates, which is a real and well documented result. Both of those hold at once, and neither is an argument for substituting an in-house opinion.
The reason to use the forward is that it is a price you can transact at. If your model says buying in March beats buying in June by two percent, and the forward market will sell you the March rate for June value, then that two percent is available without holding any stock at all. A house view cannot be transacted, and a plan built on one has no way of being shown wrong until the money is gone.
Where no forward market exists, because of capital controls or a currency too thin to quote, non-deliverable forwards sometimes fill the gap and sometimes do not. The honest fallback is the interest rate differential between the two currencies used as the expected drift, with the assumption written into the model in a place a reviewer can find it and disagree with it.
Hedging instead of timing
If the exchange rate is the only reason anyone is proposing an early purchase, the cleaner instrument is to buy on the operational schedule and hedge the payable. That separates two decisions and puts each with the person holding the information: the planner decides when stock is needed, treasury decides what currency exposure the business wants to run.
The cost of a forward hedge is the forward points, which is the same interest differential from the section above, so hedging leaves the depreciation in place and converts an uncertain future cost into a known one at a price. What it buys that early purchasing cannot is the removal of currency exposure with none of the physical cost, so no warehouse, no capital tied up in goods, no shelf life consumed while you wait.
The frictions are real enough to matter. Hedging needs credit lines and sometimes margin, both of which consume the same facility capacity the business uses for working capital, so a hedging programme and a stock build compete for the same limit. Under IFRS 9 the accounting treatment turns on documentation prepared at inception, and a hedge that fails the documentation test produces P&L volatility rather than removing it. In markets where forwards are unavailable or the tenor you want is not quoted, timing is the only instrument you have, and the cost curve becomes the whole toolkit rather than a cross-check on treasury.
The limit
The honest output of this analysis is a cost curve across candidate order months, with an interval attached rather than a point.
Every input carries uncertainty. The forward rate is a price rather than a forecast. Funding cost depends on facility availability that changes. The storage step depends on what else is in the building that month, which depends on other people's ordering decisions. The demand quantity carries a forecast error of its own, and arriving early with the wrong quantity costs more than arriving late with the right one. How much the lead time distribution widens the ordering window is a statistical problem with its own treatment, and it belongs beside this analysis rather than inside it.
What the curve reliably shows is shape. In most cases it is flat across two or three months and rises steeply outside that band, which is the most useful thing it can tell you, because it means the timing decision inside the band is worth almost nothing and the buyer should place the order on whatever date is operationally convenient. When the curve is steep, it is usually steep for a reason you can name: a duty change, a shelf life constraint, a price list date. Steepness driven purely by a currency drift assumption deserves suspicion, since that assumption is the weakest input in the model.
Presenting a single confident date does damage beyond the individual order. It invites a review six months later that scores the buyer against the rate that actually happened, which is unfair in both directions and teaches the team to protect themselves by buying early and blaming the model afterwards. Mark the flat region, state the assumption that would have to change to move the recommendation, and let the buyer place the order. Where the currency move eventually has to be recovered from the selling price, that is a pass-through decision with its own machinery and its own owner.
Start here
Take the last twelve months of import purchase orders and reprice each one as though it had been placed a month earlier and a month later, using the rates that actually occurred and your actual funding cost for the period. That backtest costs a day, uses only data you already hold, and tells you whether the timing decision in your business has been worth anything at all.