Independent researchValuation Canadian taxPrimary source: SEDAR+

The tax base nobody discloses

Dollarama does not publish the undepreciated capital cost of its property, and no Canadian issuer has to. This piece backs that base out of a deferred tax balance, runs a capital cost allowance schedule against it under the rates and first year rules actually in force, and then asks the only question that matters: how much of the answer does the tax work move?

01The hypothesis

Canadian tax law does not use book depreciation, so a model that substitutes one for the other is answering a different question.

If the tax base is far below the book base, then a model using book depreciation understates the shield.

A discounted cash flow needs cash taxes, and cash taxes need a tax deduction for capital assets. The deduction available to a Canadian corporation is capital cost allowance, computed class by class on undepreciated capital cost at rates prescribed by regulation. It is not depreciation. Depreciation is a financial reporting estimate of how an asset is consumed, made by management against the useful lives it selects; capital cost allowance is a statutory entitlement that pays no attention to useful life at all. The two produce different numbers in every year and the same number only in total, and only once the asset is gone. A model that substitutes one for the other has not made a simplifying assumption so much as changed the subject.

The reason the substitution survives is that the correct figure is not published. An issuer discloses cost, accumulated depreciation and net book value. It does not disclose undepreciated capital cost, because no accounting standard asks for it and no securities regulator requires it. So the analyst who wants the real shield has to reconstruct the base from something the issuer did publish, and then has to say honestly how much of the answer rests on that reconstruction. That is the whole exercise here.

02What the filing does say

The tax base is not disclosed anywhere, so it has to be backed out of the deferred tax note, and only one reading of that note is arithmetically possible.

Note 16(b) reports a deferred income tax liability on property, plant and equipment of 739,329, against a combined federal and provincial statutory rate of 26.5 per cent that the same note reconciles from. A deferred tax liability is a rate applied to a temporary difference, so dividing back gives a temporary difference of 2,789,921. That figure has to be reconciled to a book base, and there are only two candidates.

Read the line as the owned property alone and the tax base is the net book value of 1,258,499 less the temporary difference of 2,789,921, which is negative 1,531,422. That is not a small problem. Under subsection 13(1) of the Income Tax Act a class whose undepreciated capital cost falls below nil brings the shortfall into income as recapture and the balance resets, so a negative aggregate base cannot be carried from one year end to the next. It persists here in both years presented. The reading is not merely implausible; the statute forbids the state it describes.

Read the line as including the right of use assets and the arithmetic resolves. A leased asset has no tax cost to the lessee, because the lessee deducts rent rather than claiming an allowance on property it does not own, so the whole of its carrying value is a temporary difference. Adding the right of use balance of 2,397,209 to the owned net book value gives a book base of 3,655,708, and the tax base implied is 865,787. The components table settles it: the deferred tax asset side carries a separate line for lease obligations, which is the lease liability, and that leaves the right of use asset without a line of its own. Property is the only place it can be. Stripping the right of use balance back out leaves a temporary difference on owned property of 392,712, and that is the number the schedule is built on.

03Two bases, drawn against each other

Land and work in progress leave both bases, and what remains carries a tax base worth 58.7 per cent of its book value.

Land is excluded from both bases, since it is neither depreciated nor depreciable property, and work in progress is excluded from both for the same reason on each side: nothing has begun to be written off for accounting, and no allowance is claimable until the property is available for use under subsections 13(26) to 13(32). What remains is 951,107 of depreciable Canadian property carrying a tax base of 558,395, which is 58.7 per cent of it. The gap of 392,712 is the accumulated head start the statute has given over management's own depreciation, and it is the reason the shield is larger than a book model shows.

04The classes, and the authority for each

Each of the issuer's own asset categories is mapped to a class in Schedule II, with the provision it rests on printed beside it.

Mapping a note's asset categories onto classes in Schedule II of the Income Tax Regulations is the part of this that is judgment rather than arithmetic, and it is set out in full so it can be disagreed with. The categories are the issuer's own column headings, copied verbatim from note 9, and the allocation of the temporary difference across them is in proportion to net book value.

05The first year rules, which a book model cannot see

The rate is not the whole story: three first year rules run on three different clocks and produce cliffs no depreciation schedule can have.

The rate is only part of what a class does in the year an asset arrives. Subsection 1100(2) of the Income Tax Regulations adjusts the pool before the rate is applied, and the adjustment is not one rule but several running on different clocks. Ordinary additions are halved. Property that qualifies as reaccelerated investment incentive property, meaning property acquired after 2024 and available for use before 2034 under subsection 1104(4.01), escapes the halving and takes an extra half again, so a general class claims one and a half times the ordinary amount, until that too falls away for property available for use after 2029. Class 50 is treated separately and generously: an extra nine elevenths of the addition, which against a 55 per cent rate is a complete write off in the first year, and which expires for property available for use after 2026. Class 13 sits outside all of it, because paragraph (b)(ii) of the description of C in subsection 1100(2) excludes leasehold interests from the half year base and the incentive provisions exclude them too, so a leasehold improvement takes a full year's claim in the year it is made.

Two secondary sources consulted while writing this disagreed on that last point, one saying Class 13 escapes the half year rule and one saying it does not. The regulation decides it, and the regulation is quoted above. This is worth stating because it is the ordinary condition of tax research rather than an unusual one.

The consequence is a schedule with cliffs in it that no depreciation schedule could produce. In fiscal 2027 the computer equipment class claims 54,012 against additions of 39,261, which very nearly empties the pool. The following year, with the incentive expired, the same class claims 18,088 on a larger balance. In fiscal 2031 the general classes lose their extra half and the store and warehouse equipment claim falls from 85,331 to 68,569 even though its pool grew. A model built on useful lives sees a smooth curve where the statute has a staircase.

06The test

The reconstruction is run forward over a year it was not fitted to and asked to reproduce a figure the issuer disclosed. It misses by 5.5 per cent.

A reconstruction backed out of a disclosed balance is worth nothing until it predicts something it was not fitted to. This one was fitted to the closing balance sheet, so the test is to run it from the opening one. Taking the prior year's deferred tax liability on property, backing out the same way, and applying the same classes and the same first year rules to fiscal 2026's actual additions gives an opening tax base of 452,759, additions of 194,276 and an allowance of 136,286. Deducting that, the Canadian cash lease payments and the interest that is not already inside them from Canadian EBITDA leaves taxable income of 1,555,863, and tax at the statutory rate of 412,304.

The issuer disclosed a current tax expense of 463,583, of which 27,229 is a Pillar Two top up computed on a different base, leaving 436,354 attributable to the Canadian corporate base. The model comes in 5.5 per cent below that. That is close enough to say the reconstruction is in the right region and not close enough to call it correct, and the residual has at least four plausible homes: the allocation of the temporary difference across classes in proportion to net book value, the apportionment of lease payments between Canada and Australia, permanent differences the rate reconciliation shows at negative 2.9 per cent of pre tax income, and the fact that the model claims the maximum allowance in every year when a taxpayer may claim any amount up to it. Nothing here is tuned to close that gap, because tuning it would destroy the only test the piece has.

07The cost of capital, as far as it can be derived

The cost of debt is solved from disclosed fair values and the beta is regressed from five years of weekly returns. Only the equity risk premium is still asserted.

The notes disclose the coupon, the maturity and the fair value of every tranche outstanding, and disclose an effective interest rate for none of them. The harvest left those fields empty rather than substituting coupons for them, so the cost of debt is solved rather than read: each tranche's fair value is the price, and the discount rate that produces it is the yield. Weighted by fair value that gives 3.170 per cent, which is below the 3.642 per cent weighted average coupon for an arithmetic reason rather than an interpretive one: the notes trade above par in aggregate, 2,648,846 of fair value against 2,600,000 of face, and a bond above par yields less than it pays.

Capital structure is taken at market: 50,082,542 of equity against 2,625,121 of borrowings, which is 95.0 per cent equity. Lease liabilities are deliberately not treated as debt here, and the reason is a tax reason. Under IFRS 16 the reported EBITDA excludes lease costs entirely, while the Income Tax Act allows the lessee the full cash lease payment as a deduction. Treating leases as debt would require adding a lease discount rate the filing does not disclose, and would then require the payment to be split between principal and interest, a split the issuer explicitly does not make. Treating them as an operating cost keeps the cash flow and the tax computation on the same footing and needs nothing that is not disclosed. The cost of equity is where the filing runs out. A risk free rate of 3.40 per cent comes from the Government of Canada ten year yield, and a beta of 0.42 is regressed here from 261 weekly returns against the S&P/TSX Composite over the five years to 2026-01-26, on the series cached in content/market-data.json so it can be recomputed. That leaves the equity risk premium of 5.50 per cent as the one number in the whole chain with no source at all, and together they give a cost of equity of 5.726 per cent and a weighted average cost of capital of 5.56 per cent.

The coefficient should not be quoted on its own, because the regression that produced it explains 7.6 per cent of the variance. Dollarama's weekly returns move at 19.5 per cent annualised against 12.7 per cent for the index, and almost none of that movement is the index. A capital asset pricing model built on this beta is a weak instrument whatever number it returns: at the bottom of the interval the cost of equity is 4.74 per cent and at the top it is 6.71 per cent, a spread wider than the entire tax question this piece is about. Section 09 therefore turns the model around and reports what discount rate the market is already using, which needs no beta at all.

08Why the range is not averaged

The two anchors are two readings of a line the issuer does not split, so the space between them is left empty rather than averaged.

The two anchors are not a bull case and a bear case, and they are not two sentiments about the same facts. They are two readings of a disclosure the issuer does not make. Capital spending of 248,747 thousand appears as a single line in the investing section of the cash flow statement, and nothing anywhere in the statements divides it between the spending required to keep 1,687 leased Canadian stores trading and the spending that opens new ones. Both readings are internally coherent. On the first, the programme is expansion, so once it stops the spending falls to replacement plus what the terminal growth rate itself requires, and the growth it bought is real, which supports terminal growth of 3.0 per cent. On the second, the programme is what standing still costs, so it never falls and there is no separable growth capital to capitalise, which supports 2.0 per cent.

The midpoint of $125.14 and $187.16 would be a number asserting that roughly half the capital programme is growth. No disclosure supports that proposition, and inventing it would convert an honest absence of information into a false precision. The gap is drawn as a gap.

09What the price requires

Both anchors are compared to the traded price by inverting the model rather than by tuning an input until it agrees.

The traded price of $183.50 falls between the two anchors, and that is a weaker result than it looks. It does not mean the model agrees with the market. It means the fork is wide enough to contain the market, which is a statement about the width of the fork. Inverting the model says it more precisely: the expansion reading reaches $183.50 at a weighted average cost of capital of 5.61 per cent against the 5.56 per cent derived here, a difference of five hundredths of a point, or at the derived rate with terminal growth of 2.95 per cent against the 3.0 per cent assumed. The maintenance reading needs 4.45 per cent, or terminal growth of 3.18 per cent against the 2.0 per cent assumed. So on one reading of the capital programme the market and this model are within rounding of each other, and on the other they are more than a point of discount rate apart, and the filing does not say which reading is right.

That is the finding, and it cuts against the piece's own subject. The capital cost allowance work is correct, checkable and worth doing, and it is worth $1.78 to $2.27 a share on a stock trading near $184. The terminal value is 90 per cent of enterprise value on the expansion reading, and at a discount rate of 5.56 per cent against terminal growth of 3.0 per cent the spread that capitalises it is barely two and a half points, so the terminal value is hypersensitive by construction. Any honest account of where the answer comes from has to lead with those two facts rather than with the tax schedule. A valuation that advertised the tax insight without them would be selling the rigorous part to distract from the load bearing part.

10The base year, and what was done to it

Fiscal 2026 is a 52 week year against a 53 week comparative and consolidates Australia for the first time, so the base year needs work before it can be forecast.

Fiscal 2026 is not comparable to its own comparative without work. It ran 52 weeks against 53 in the prior year, and it consolidates an Australian segment from 22 July 2025 that the prior year does not contain. Face sales growth of 13.1 per cent therefore overstates nothing and understates a great deal. Removing the Australian segment and putting both years on a per week basis gives organic Canadian growth of 8.09 per cent, and that is the rate the forecast fades from.

The valuation is built on the Canadian segment alone, because capital cost allowance is a deduction under a Canadian statute and a shield computed on consolidated property would be a category error. The Australian property arrived through the business combination and appears in note 9 only in the movement lines, which is what makes the carve out clean: removing them removes 71,572 of cost from the base. Reported operating income is also not a retail figure, because it is struck after adding 191,536 of equity accounted earnings from the Dollarcity group, which trades in El Salvador, Guatemala, Colombia, Peru and Mexico. Those earnings are not Canadian cash flow and are not taxable on this base, so they are removed from the forecast and the investment is carried at its balance sheet amount of 1,285,105 among the non operating assets instead.

11What would make this wrong

Every joint in the model that could be wrong, and how much each one is worth.

The reconstruction of the tax base is the load bearing assumption and it is indirect. It rests on reading one line of a components table as containing the right of use assets, and although the alternative reading is arithmetically impossible, a third possibility exists that neither this piece nor the disclosure can rule out: that the line nets something else against the property difference. If the base is wrong the whole schedule is wrong, and the test in section 06 constrains the error to roughly 5.5 per cent of one year's tax rather than eliminating it.

The allocation of the temporary difference across classes in proportion to net book value is a convenience with no evidence behind it, and it is worth measuring rather than worrying about. Pushing the whole difference onto the fastest pools gives $187.14 per share and pushing it onto the slowest gives $187.19, against $187.16 published. Every allocation that is possible lies between those two, so the convenience is worth $0.05 per share. The aggregate base is fixed by the deferred tax balance, the split moves only timing, and at a discount rate of 5.56 per cent the timing is nearly worthless. This was named as the weakest joint in the model in an earlier draft. It is the weakest joint and it costs almost nothing, which is a different and more useful thing to know.

The Class 13 period of 8 years is derived from the right of use balance against right of use depreciation, which is a proxy for the average remaining lease term and not a disclosure of it. The lease payment split between Canada and Australia is estimated from the lease liability that came in on acquisition and the part of the year it was consolidated. The additions figure includes Australian capital spending after 22 July 2025 that the note does not separate. The model claims the maximum allowance every year, where a taxpayer may claim less. And the share price is the only figure in the whole chain that does not come from the filing, taken from a single market source and not cross checked.

None of these is fatal on its own and none of them is hidden. The one that would change the conclusion is none of them: it is the terminal growth rate, which carries 90 per cent of the answer.