🧮 PAC allocator

The PAC allocator answers one question:

With the money available for this contribution cycle, which purchases bring my allocation as close as possible to its target?

Open Tools from the sidebar and click the PAC allocator card: it opens a guided planner. You describe a scenario step by step, press Calculate plan, and get a purchase plan to evaluate. It is a simulation: nothing is bought and no order is sent to a Broker.

🗺️ Using the planner

The planner guides you through these steps, in order. The FX step appears only when the scenario needs exchange rates.

🎬 Scenario

You plan a new investment, a "pure PAC": the calculation starts from an empty portfolio, so what you already hold is not counted, and it splits the cash you choose among the Assets, as close as possible to the target weights.

Here you set the Valuation currency, in which amounts are compared and summarized. At first it is the Default Currency of your preferences. Money stays in its own currency: every exchange needed appears in the plan.

The reference date is always today: the planner sets it again before every copy from LibreFolio and before every calculation.

💰 Liquidity

Where does the money come from? You can combine several sources:

• From your Brokers: cash already on your Brokers in LibreFolio. You choose how much of it to use.
• New contribution: new money you add, such as the PAC instalment.
• External account: a balance on an account not registered in LibreFolio, for example at your bank. You declare how much is there and how much to use. Nothing is bought from an external account: its money is sent to a Broker.

The Broker that receives a contribution or an external account's money is chosen in the Brokers step. Each amount stays in its own currency.

🏦 Brokers

Add the Brokers where the plan may buy: Choose existing Broker copies one of yours, and Manual Broker adds one by hand. Either way they are scenario data: the Broker itself is never changed. Fees and order modes are not recorded in LibreFolio, which is why you set them here.

For each currency you buy in, set:

• The order mode: By number of units or By amount.

• The Increment \(\Delta\): every proposed order is a whole multiple of it. With \(q\) the size of an order, in units or as an amount:

  \[ q = k\,\Delta, \qquad k = 0, 1, 2, \dots \]

  By number of units, \(\Delta\) is a whole number: \(\Delta = 1\) means whole units only. By amount, \(\Delta\) is the smallest amount you can enter, such as \(\Delta = 0.01\), and the units you get can be fractions of a unit.

• The Purchase fee: a Fixed part \(f\) plus a Percentage \(r\) of the order amount \(A\), price margin included. The Minimum \(f_{\min}\) and the Maximum \(f_{\max}\) limit the percentage part only, and an empty Maximum sets no upper limit:

  \[ \text{fee} = f + \min\big(\max(r\,A,\ f_{\min}),\ f_{\max}\big) \]

  The fee is charged on every order, and no order means no fee. It is paid from your liquidity and is not invested. For example, with \(r = 0.19\%\), \(f_{\min} = 1.50\), \(f_{\max} = 18\) and no fixed part, an order of \(500\) pays \(1.50\), one of \(2{,}000\) pays \(3.80\), and one of \(20{,}000\) pays \(18\).

Each Broker also has two settings of its own:

• Usable liquidity: the money the plan may use to buy there, that is the Broker's own cash plus the contributions and other accounts you allow. All of them are allowed at first: you can exclude a source, cap how much of it is used there, and give it a Priority (\(0\) = preferred).
• Currency conversion: You convert before buying or The Broker converts when you buy. Both convert at the rate of the FX step minus the spread, so the calculation is the same either way: only the way the plan shows the conversion changes.

Not supported yet: a tax regime, losses, and sell fees for a Broker.

💼 Assets

Add the Assets the plan may buy, in any mix:

• search the Assets registered in LibreFolio;
• add Your Assets: those with an open position today in your Brokers, added as rows only, without quantities;
• add a Manual Asset.

Each price is either:

• Auto: the latest price stored in LibreFolio, read again just before the calculation. No data provider is called.
• Manual: your own price, used as typed and never read again.

A missing price stays in the draft: the calculation asks for it.

The composition by country, sector, and type, copied from the Asset or typed, feeds only the exposure maps and bars of the result, not the calculation.

🔀 Routing

For each Broker, choose which Assets it may buy: Allow all, Exclude all, or click each one.

Then, only where you need them, set limits on each route. They are measured like that Broker's orders, in units or as an amount. With \(q\) the purchase of the Asset on that Broker:

• Minimum purchase \(q_{\min}\): if the plan buys there, it buys at least this much, so \(q = 0\) or \(q \ge q_{\min}\).
• Required purchase \(q_{\text{req}}\): bought whatever happens, \(q \ge q_{\text{req}}\). If the resources are not enough, the plan becomes impossible.
• Maximum purchase \(q_{\max}\): the most the plan may buy there, \(q \le q_{\max}\).

Two more settings shape how the plan buys there:

• Priority (\(0\) = preferred): it only breaks ties between plans equally close to the target.
• Price margin \(m\): each purchase is counted at \(p\,(1 + m)\) instead of the price \(p\) of one unit, to cover a price that rises before the order. A margin of \(0.5\%\) on a price of \(100\) counts \(100.50\). The difference is a reserve, not invested. By amount, an order of \(A\) buys \(A / \big(p\,(1 + m)\big)\) units.

Empty fields restrict nothing.

⚖️ Targets

Here you split the money you invest now among the Assets: one target \(w_i\) per Asset, in percent, decimals allowed. To calculate, the targets must add up to exactly \(100\%\):

\[ \sum_i w_i = 100\% \]

Two actions help you get there:

• Balance all rescales every target, keeping their ratios:

  \[ w_i' = \frac{w_i}{\sum_j w_j} \cdot 100\% \]

  If they are all \(0\), every Asset gets an equal part. Rounding keeps the total at exactly \(100\%\). To change only some rows, Balance to 100% on a row gives that Asset what is missing, or takes away the excess, and Balance the selected rows to 100% rescales only the rows you tick.

• Copy current distribution reads how much each of these Assets weighs today in the Brokers you tick, counting only these Assets and not cash. With \(H_i\) the market value of Asset \(i\) held there today:

  \[ w_i = \frac{H_i}{\sum_j H_j} \cdot 100\% \]

  The weights are rounded to \(0.01\) percentage points and still add up to exactly \(100\%\), and an Asset you entered by hand gets \(0\). These are weights only, a base to edit and not advice: you see them before using them, and a target you changed is not overwritten without your confirmation.

💱 FX (only when needed)

This step appears only when the scenario needs exchange rates. It holds one rate \(x\) per currency pair in play, each Auto or Manual:

• Auto: the latest rate stored in LibreFolio, read when the step opens and again just before the calculation. No data provider is called.
• Manual: your own rate.

It also holds one Conversion spread \(s\): a percentage of every converted amount, kept as a margin for a change in the official rate or for Broker fees. Converting an amount \(D\) at the rate \(x\) (units received for each unit converted) gives

\[ D \cdot x\,(1 - s) \]

instead of \(D \cdot x\). Valuation, which compares and sums amounts in the valuation currency, uses the official rate \(x\) without spread; conversions use the rate with the spread, \(x\,(1 - s)\).

The rates must also agree with each other, so that no conversion creates value: with cash in CHF, an Asset in USD and valuation in EUR, for example, one CHF converted into USD after the spread may not be worth more in EUR than one CHF,

\[ x_{\text{CHF} \to \text{USD}}\,(1 - s)\,x_{\text{USD} \to \text{EUR}} \le x_{\text{CHF} \to \text{EUR}} \]

and the same holds for every conversion the plan may need between two currencies other than the valuation currency. Otherwise the calculation does not start and the outcome is Input not valid (see Reading the result): the cause can be a Manual rate or Auto rates from different days or sources, so align the rates or set a Conversion spread that covers the difference. Only a gap small enough to come from storing rates with ten decimals is tolerated: the conversion then uses the rate through the valuation currency when that is lower,

\[ \min\left(x_{\text{CHF} \to \text{USD}}\,(1 - s),\; \frac{x_{\text{CHF} \to \text{EUR}}}{x_{\text{USD} \to \text{EUR}}}\right) \]

so it still creates no value. When the rates agree, this rate is the usual \(x\,(1 - s)\). Your rates are not changed: each conversion of the plan shows the rate it starts from as spot and the rate it uses as effective.

Every amount the plan books is rounded once, on its exact final value, to the smallest unit of its currency (the cent for EUR or USD), and always against the plan: amounts it receives (what a conversion delivers) are rounded down, amounts it pays (the cost of an order and its fee) are rounded up. So rounding never improves a plan, splitting a conversion into smaller ones gains nothing, and the rounding part of Not invested is never negative. The Rounding column of Balances per Broker and currency shows how much of each row comes from rounding: ≈ marks a figure shown rounded, as when the exact difference has no finite decimal form (for example after converting USD → EUR at \(1/1.085\)), and a difference below the smallest unit keeps the digits it needs, such as ≈ −0.0022 rather than ≈ −0.00.

When LibreFolio has no rate for a pair, the step offers Add the pair or Download the rates: each opens the matching window of the FX page, and nothing is added or downloaded until you confirm there.

Not supported yet: a different exchange rate or spread per Broker, a safety margin on the rate, a conversion fee on top of the spread, and conversions in more than one step (for example EUR → USD → CHF).

🧠 Strategy

The strategy is Proportional: purchases only, it sells nothing.

Among all the purchase plans that respect your settings, it picks the best one with a cascade of criteria, each deciding only among the plans still tied on the ones above it:

1. Closeness to the targets (L2 distance): the smallest distance

   \[ D = \sum_i \big(V_i - w_i\,R\big)^2 \]

   where \(V_i\) is the value of Asset \(i\) after the plan, \(w_i\) its target, and \(R\) the Base of the targets: the cash you chose that can reach a Broker where it can buy (in a PAC nothing is already invested). So \(w_i\,R\) is the ideal value of Asset \(i\). Values are in the valuation currency, at the quote price, without fees or price margin. Squaring makes big gaps weigh more, and \(D\) is measured in squared money, for example EUR².

2. Uninvested money: the least money left out of the Assets, \(R - \sum_i V_i\).

3. Broker and source priority: the smallest sum of the Priority numbers of the orders and cash sources the plan uses.

4. Explicit costs (fees, spread, margin): the lowest total of fees, conversion spread, and price margin, in the valuation currency.

5. Number of orders: the fewest orders.

A fixed order of Assets and Brokers settles any final tie, so a search that completes always gives the same plan for the same data; a search stopped by a time or node limit can give a different plan on a slower or busier machine.

✅ Review

A last check before the calculation. It lists the complete copy that will be sent (the backend receives this copy, and only this), flags the fields still to complete, and offers Calculate plan.

📋 Copied or typed values

The planner works without any registered Broker or Asset: everything can be typed. When you would rather start from your LibreFolio data, you copy it with an explicit action: Choose existing Broker, From your Brokers, the Asset search or Your Assets, and Copy current distribution. In the FX step, an Auto rate is read from LibreFolio as soon as the step opens.

Each value shows where it came from (Copied, Manual, or Modified), and a copy also shows its date. A copy does not follow its source while you edit, and a copy you modified can be restored.

When you press Calculate plan, the planner first reads again from LibreFolio the prices, rates, and balances you copied and have not changed. Values you typed or changed stay as they are. If that read fails, nothing is calculated: you can Try again, or Calculate with the copied data to use the earlier copies.

⏳ While it calculates

While the request is active, the planner shows Calculation in progress and the configuration is locked. Stop waiting stops only the wait: the server may finish anyway, and that answer is discarded.

If you change the draft after a result, a Result not up to date banner appears: the previous values can still be consulted, but they no longer describe the current draft. From the banner, Back to Review returns to the last step, and Discard the previous result removes the old result and takes you there too.

📊 Reading the result

A calculated plan opens with a header: the scenario date, the valuation currency, and the draft revision, then a row of badges. Each badge explains itself when you point at it, tap it, or focus it. A plan also shows its L2 distance from the targets (the \(D\) of the Strategy step), how much is Not invested, and how many notes the calculation left; the notes are listed just below, under Notes on the calculation. Edit configuration takes you back to the Review step, and Calculate new plan runs the calculation again.

A calculation ends with one of these outcomes:

Outcome	What it means
Plan available	A plan that respects every constraint. A second badge says what it is worth. Proven optimal: the solver proved, within its small calculation margin (which grows with the amounts), that no better plan exists, objective by objective in the order of the Strategy step, and the exact check of the plan matches its numbers. Optimality not proven: the solver reached its time or node limit, or its own numbers do not match the exact check of the plan; it is then the best plan found, but it is not proven to be the best.
No operation	With the cash you chose, no purchase meets the constraints: the plan is to do nothing. No order, no currency exchange, and no funding.
Infeasible with these constraints	Marked Infeasibility proven: no combination meets every hard constraint together. The constraints involved are listed (each Required purchase, and the liquidity that can reach the Brokers), each with a link to the step where you can change it. The planner does not choose which one to relax, and no partial plan is shown as valid.
No plan within the limits	The solver found no plan before reaching its limit. This is not a proof that none exists. A lighter search helps: fewer Assets or routes, a lower Maximum purchase, or a larger Increment.
More data needed	Something is missing. Your draft is intact, and no Asset is removed silently: add the missing fact or remove the Asset yourself.
Input not valid	Some of the data are not valid.
Scenario not supported	Not an error in your data: the tool does not handle this case yet.

Every plan shown has been checked again in exact decimal arithmetic, independently of the solver: that is the Verified in Decimal badge. A last badge tells how the search ended: Completed (the solver ended its search by itself), Time limit, or Node limit. When a plan was found but the search stopped at a limit, a notice adds that a better plan may exist; the result stays complete and can be consulted. If rounding to the smallest unit of a currency leaves a Broker slightly short, a notice says how much more cash that Broker needs to execute the plan.

The last three outcomes mean that the calculation could not start on your data: nothing is calculated, and the problems found are listed, with a link to the step concerned when there is one. Details adds the backend code of each problem.

Platform errors, such as a timeout, a full queue, or a crashed worker, are not financial conclusions about your scenario: your draft stays intact, so try again later. If the tool is unavailable, see Settings → About → Plugin diagnostics.

🗂️ How a plan is laid out

Below the outcome, a plan is laid out in this order:

1. Key figures: the plan in a few numbers, Base of the targets, Invested after, Not invested, Cash chosen, Costs, and Orders, each with a ? that explains it and, below the value, the parts it is made of. Beside them, the Calculation box shows how long the optimizer worked, the time it was allowed for each objective, and how many objectives it closed.
2. Allocation per Asset: for each Asset, its target share next to its share after the plan, with its ideal value \(w_i\,R\), its value after the plan \(V_i\), the gap from the ideal \(V_i - w_i\,R\), and the value bought.
3. Operational plan: the steps to follow, in order. First the cash (Available cash, Transfer, Deposit), then the currency exchanges you make yourself, then one table of orders per Broker, with the Instruction to enter at the Broker, the Price, the Order amount, and the Fee. A conversion that the Broker makes by itself when you buy has no number: it appears above that Broker's orders as an Automatic conversion. This section is shown only when the plan has something to do.
4. Exposures – ideal vs actual: maps and bars by country, type, and sector, built from the compositions of the Assets. The charts do not change the plan.
5. Balances per Broker and currency: how each cash balance moves, and what is left.
6. Proof and solver: how the outcome was established. The outcome, proof, and stop badges, the exact value of each objective, the solver stages, and the backend timings.

Allocation per Asset and Operational plan start open; the other sections open on request, and Expand all / Collapse all opens or closes them together. Without a plan (Infeasible with these constraints or No plan within the limits), Proof and solver is the only section, and the key figures show only the Calculation box.

Click an order, or its Detail button, to open its detail: the instruction, the economic quantity, the prices used (source price, mid price, and charge price with the price margin), the cash debit and the fee, the conversions that pay for it, and the priority, limit, and minimums of its route. Show provenance lists where each value came from, Copied or Manual, with the date and time of the copy or of your entry.

Proof and timings, at the top of the result, opens Proof and solver and scrolls to it.

Every figure comes from the backend accounting; the interface adds nothing up. With privacy mode on (see User preferences), the result hides what would tell how much you own: every amount (key figures, values, balances, order amounts, fees and costs, and the objective and solver values measured in money, the L2 distance included), every quantity, and the purchase limits of a route (Minimum purchase, Required purchase, Maximum purchase). The amounts in the problems listed when a calculation could not start are hidden too. Market prices, exchange rates, percentages (targets, shares, exposure weights, spreads, and margins), increments, priorities, counts, and dates stay visible, because they say nothing about how much you own.

🧮 What the calculation engine does

The engine behind this tool plans purchases. Given a set of Assets with their prices, the Brokers and buy routes that can be used to reach them, the cash and contributions available, and one target weight per Asset, it searches for the combination of purchases whose resulting allocation sits as close as possible to those targets.

Three properties are worth knowing, because they shape what the planner can promise:

• It buys in each Broker's increments. Whole units or amounts, always in multiples of the increment you set: increments, fees, and the currencies involved are part of the problem it solves, not a rounding step applied afterwards.
• Its published numbers come from exact arithmetic. Every candidate plan is re-checked exactly before anything is shown. A plan that fails that check is never published.
• It never passes off an unproven plan as optimal. A plan stopped by a time or node limit is shown as the best one found and marked Optimality not proven. So is a plan whose exact check does not match the solver's own numbers, even when the search ended by itself. When no plan is found, it says so instead of guessing. With amounts of around ten billion units of a currency or more, the solver's calculations can lose precision: the tool may then stop with an error, or mark the plan Optimality not proven.

A completed calculation therefore reports one of the small set of honest outcomes listed in Reading the result.

The engine plans purchases only. It does not plan sales.

🎯 How its target is meant to be read

The PAC allocator distributes the liquidity available now — existing cash plus new contributions. Its targets describe how that money should be allocated; they do not describe the final mix of a portfolio you already hold.

The calculation does not take the positions you already hold as an input at all, so changing them cannot change what this tool plans. Copy current distribution only turns them into target weights for you to edit, and those weights do not follow later changes.

🚫 What this tool never does

The planner and the calculation behind it stay within the Tools platform contract:

• it does not place orders, execute trades, or contact a Broker;
• it does not write to your Assets, Brokers, or transactions;
• it does not read your portfolio by itself: the calculation receives only the scenario sent from Review, and the planner reads your LibreFolio data only when you copy something or open the FX step (for its Auto rates), then again just before a calculation for the copied values you have not changed;
• its result is a plan to evaluate, not advice and not an instruction.

🔒 Your financial data

The calculation runs on the scenario submitted with the request. It is not handed a live portfolio, a database connection, or your signed-in session, and it produces nothing that is stored.

Your draft lives only in the open page: it is saved neither on the server nor in the browser. Leaving the planner, or reloading or closing the tab, with a draft in progress asks you to confirm first. Signing out or switching account discards the draft without asking.

As always, avoid pasting real portfolio values, Broker exports, or account identifiers into examples or support messages.

🔗 Related

• Tools overview
• User preferences, including privacy mode