Every number, accounted for

How each figure in Dollar Odyssey is built, what it rests on, and where it can be wrong. Written for readers who want to check our work.

Last updated: October 2026

Principles

Every figure in Dollar Odyssey follows the same small set of rules. They are what make a percentage on one page comparable with a percentage on another.

  • Value-weighted, always. A holding counts in proportion to its current value in your display currency, never by the number of lines it occupies.
  • One base per page. Every share on an analysis page is a share of the same total: the value of the positions inside the asset-type filter you selected. Shares on a page therefore add up.
  • Missing data is shown, never guessed. Value we cannot attribute to a country or a sector lands in an explicit no data bucket. It is never spread silently across the others, so a gap can never inflate an exposure.
  • Cash is shown, not judged. Cash appears in every breakdown, but it is left out of concentration verdicts, otherwise a cautious portfolio would read as "concentrated in cash".
  • Live results, curated reference data. Your exposures, scores and simulations are recomputed from your current holdings, prices and FX rates. The reference data behind them (classifications, country profiles, macro sensitivities) is researched, reviewed and refreshed on its own cycle.

Why you can trust the numbers

A number is only as good as the process behind it. These controls run continuously, whether or not anyone is looking.

  • One security, one record. Listings of the same security on several exchanges, in several currencies or as depositary receipts are resolved to a single asset by its international identifier, so nothing is counted twice.
  • Provenance on every field. Each classification weight records where it came from (provider data, fund holdings or our own research) and when it was last reviewed.
  • Research beats defaults. Where our analysts have studied a company, their view takes precedence over any provider default, and it is never overwritten by a later data reload.
  • Strict validation. Every load checks that classification weights are positive and sum to 100%. Unknown codes stop the load instead of falling silently into Unclassified. Inactive and delisted lines are removed.
  • Independent spot checks. Classifications are re-checked in sampled reviews against company filings, and corrections are stored as permanent, replayable rules.
  • Coverage in plain sight. Every analysis page states how much of your portfolio it could classify and lists the holdings behind any gap. Profit and loss tells you when a cost basis is missing instead of assuming one.
  • Reproducible simulations. A Monte Carlo run is seeded: the same plan and seed replay exactly the same result.
  • Computed on the server. Scores and scenarios are computed server-side from the same positions you see, not from a separate copy.

Data sources

DataSource
Listed securities (stocks, ETFs, funds, bonds)Several licensed market-data providers covering dozens of exchanges worldwide, merged field by field under an explicit precedence rule
Fund constituentsPublished fund holdings, taken from the most complete source available for each fund
Company geography and business mixOur own research on company disclosures: annual reports, regulatory and exchange filings, investor presentations, sustainability reports
Crypto-assetsMarket data providers and protocol documentation
FX rates and metalsReference rates from market-data providers
Regions and country codesUN M49 standard (UN Statistics Division), ISO 3166-1, ISO 4217
Country risk inputsWorld Bank Worldwide Governance Indicators, IMF World Economic Outlook database, IMF AREAER, published sovereign credit ratings, index-provider market classifications
Sector taxonomy and macro sensitivitiesOur own, maintained by our research team

When providers disagree, the rule that decides is fixed in advance and applied per field, never "whichever file arrived last". Every field keeps a record of the source that won.

How assets are classified by geography

Geography measures economic dependency, not legal address. A company listed in Kuala Lumpur that designs chips sold mostly in the United States is, economically, mostly a US exposure. Listing exchange, headquarters, incorporation and registered office are never used on their own.

Companies

Each researched company receives a country map built from three pillars of evidence:

  1. Revenue and customers: where demand originates (revenue by country or by segment, customer location, end markets, export destinations). This pillar weighs the most.
  2. Sourcing and supply chain: where suppliers, raw materials, components and contract manufacturers are.
  3. Operations and assets: where factories, offices, data centres, employees and fixed assets are.

For each country cc, the pillar shares xp,cx_{p,c} are blended with fixed pillar weights αp\alpha_p and normalised so the map sums to 100%:

gc=∑pαp xp,c∑c′∑pαp xp,c′g_c = \frac{\sum_{p} \alpha_p \, x_{p,c}}{\sum_{c'} \sum_{p} \alpha_p \, x_{p,c'}}

A pillar without evidence is not read as zero exposure. Its weight is redistributed proportionally over the pillars that do have evidence:

αp′=αp∑q∈availableαq\alpha'_p = \frac{\alpha_p}{\sum_{q \in \text{available}} \alpha_q}

Evidence is taken in a fixed order of authority: latest annual report, regulatory filings, investor presentations, sustainability report, company website, facility and subsidiary data, then credible databases and press. When sources conflict, the most recent official disclosure wins. Countries are used whenever they are reliably disclosed; nothing is invented at country level. Exposure disclosed only by region, or not at all, stays in the no country data bucket in the app rather than being forced onto a country. Companies not yet researched carry their issuer country as a provisional value.

Illustrative example (not real figures): a manufacturer selling mainly in the United States, sourcing mainly in China and producing mainly in Malaysia ends up with a blended map in which each of the three countries carries a material share, instead of 100% in its listing country.

Funds and ETFs

A fund is classified by look-through: its map is the weighted average of its constituents' maps, where ωj\omega_j is constituent jj's weight in fund FF:

wF,c=∑j∈Fωj wj,cw_{F,c} = \sum_{j \in F} \omega_j \, w_{j,c}

Funds of funds are expanded level by level down to the underlying securities, multiplying weights along the chain. Look-through is used only when enough of the fund resolves to securities we know. Below that, the fund's published country breakdown is used, and failing that its domicile. A top-10 snapshot is not a fund.

Other assets

Cash and currency holdings follow the issuing country (split evenly for a currency shared by several countries) and property follows its location. Crypto-assets and precious metals have no meaningful country and are reported as such.

Regions

Countries roll up to subregions and regions following the UN M49 standard of the UN Statistics Division, so "Western Europe" or "South-eastern Asia" mean exactly what they mean in official statistics.

How assets are classified by sector

Sector follows the same idea as geography: economic substance, not a single label.

Our taxonomy

We maintain our own three-layer taxonomy (13 sectors, 61 industries and 183 sub-industries, plus Unclassified), designed for how private investors actually perceive companies. Internet platforms sit in Technology, payments in Banking & Finance, and Digital Assets is a sector in its own right. Every node has a stable hierarchical key, so a holding can be read at any depth and history stays consistent.

Weighted attribution

A company has no primary sector. It carries weights across sub-industries that sum to 100%, and every weight is counted:

∑kwi,k=1,wi,k≥0\sum_{k} w_{i,k} = 1, \qquad w_{i,k} \ge 0

Analysts set these weights from three inputs, applied the same way to every company:

  1. Business segments: what the company actually does, segment by segment, from its reporting.
  2. Corporate purpose: its stated objective and strategic direction.
  3. Current profit: where today's profit comes from.

Illustrative example (not real figures): an electric-vehicle maker can carry most of its weight in vehicle manufacturing and a meaningful share in Technology.

Rules that keep it honest

  • Defense carries a weight only when military contracts reach a material share of revenue, weighted by that share from segment reporting. Public contract-award data helps find candidates; filings set the figure.
  • Activities investors commonly exclude (alcohol, tobacco, casinos and betting, coal, defense, civilian firearms, cannabis) each have their own leaf, so they can always be isolated precisely.
  • AI is identified at company level: a company's AI activity is split from its other activity by weight, never counted twice.
  • Share classes and depositary receipts carry exactly the same weights as the main line.
  • Funds use the same look-through as geography, on our taxonomy.
  • Crypto-assets are classified by what the protocol does (base layer, application, stablecoin by type of backing, and so on).

Exposure and concentration

Exposure

Position ii with value ViV_i and classification map wi,⋅w_{i,\cdot} contributes to bucket cc (a country or a sector):

Ec=∑iVi wi,cmax⁡ ⁣(1,∑kwi,k)E_c = \sum_i V_i \, \frac{w_{i,c}}{\max\!\left(1, \sum_k w_{i,k}\right)}

A map that sums above 100% (a data-format error) is scaled back. A map that sums below 100% keeps its true coverage, and the remainder goes to the no data bucket:

E∅=∑iVi(1−min⁡ ⁣(1,∑kwi,k))E_{\varnothing} = \sum_i V_i \left(1 - \min\!\left(1, \sum_k w_{i,k}\right)\right)

The share shown on screen is sc=Ec/∑iVis_c = E_c / \sum_i V_i, so classified and unclassified value always add up to the whole page. The coverage line at the bottom of each analysis page reports E∅E_{\varnothing} and the holdings behind it.

Concentration

Concentration is judged on holdings that are classified and not cash. It reads two figures, the largest bucket S1S_1 and the three largest combined S3S_3, and classifies the portfolio as diversified, moderate or high. A single bucket is flagged above 20% of investments for a geographic subregion and above 25% for a sector, and the flag grows in severity as the share rises.

Geography risk scores

Country profiles

Every country is profiled on six dimensions, each scored from 0 (lowest risk) to 5 (highest), plus an overall score:

DimensionWhat it captures
PoliticalStability, governance, institutional quality
CurrencyExchange-rate regime, volatility, convertibility
InflationLevel and persistence of price growth
RegulatoryRule of law, regulatory quality, investor protection
MarketDepth, liquidity and maturity of the capital market
SovereignPublic debt sustainability and sovereign credit standing

Profiles are built by our analysts from public indicators (World Bank Worldwide Governance Indicators, IMF World Economic Outlook data, IMF AREAER exchange arrangements, sovereign credit ratings) and reviewed periodically. Each profile also carries a market type (developed, emerging, frontier) and its key risk themes.

Your portfolio's score

For each dimension dd, your score is the exposure-weighted average over the set CC of held countries that have a profile:

Rd=∑c∈CEc rc,d∑c∈CEcR_d = \frac{\sum_{c \in C} E_c \, r_{c,d}}{\sum_{c \in C} E_c}

Countries without a profile are not scored as zero. They are left out, and the covered share is reported next to the score:

κ=∑c∈CEc∑c∈HEc\kappa = \frac{\sum_{c \in C} E_c}{\sum_{c \in H} E_c}

where HH is every country you hold. Subregion scores are computed the same way from the countries you actually hold in them, never from a generic regional average. The risk themes shown are the most exposure-weighted themes across your countries.

Sector macro sensitivity and scenarios

Industry sensitivities

Each industry is scored by our analysts on how it tends to react to 13 macro factors: interest rates, inflation, economic growth, credit conditions, the US dollar, oil prices, commodity prices, consumer confidence, technology adoption, regulation, geopolitical risk, earnings growth and market risk appetite.

Rises and falls are scored separately, each on a scale from −5 to +5, because markets rarely react symmetrically: an industry can suffer a lot when rates rise and gain little when they fall.

Your portfolio's exposure

For each factor movement kk, your exposure is the value-weighted average of the scores gj,kg_{j,k} of the industries jj you hold, with uju_j the value invested in industry jj. Unclassified value carries no score and is left out.

γk=∑juj gj,k∑juj\gamma_k = \frac{\sum_j u_j \, g_{j,k}}{\sum_j u_j}

The net sensitivity to factor ff compares the rising and falling scores:

nf=γf↑−γf↓2n_f = \frac{\gamma_{f\uparrow} - \gamma_{f\downarrow}}{2}

A positive nfn_f means the holdings tend to benefit when the factor rises; a negative one, when it falls. Small values read as no clear lean, larger ones as a weak, medium or strong lean. The factor grid shows the factors that apply to the largest part of your portfolio.

Scenarios

A scenario is a predefined combination of factor movements with weights cs,kc_{s,k} (for example, rates rising while growth slows). Its impact on your portfolio is:

Is=∑kγk cs,kI_s = \sum_k \gamma_k \, c_{s,k}

The sign gives the direction (tailwind or headwind) and the size gives the magnitude. To show why, the impact is broken down by industry and by holding. Each contributor's share is its part of the gross impact, ∣Is,j∣/∑j′∣Is,j′∣|I_{s,j}| / \sum_{j'} |I_{s,j'}|, so offsetting positions are visible instead of cancelling out unseen.

These are structured sensitivities, not forecasts: they describe direction and relative size, not a price target.

Profit and loss

The acquisition price is recorded per unit. The cost basis of position ii with quantity qiq_i is therefore:

Costi=piacq×qi\text{Cost}_i = p_i^{\text{acq}} \times q_i

expressed in your display currency. Unrealised profit and loss compares it with the current value:

P/Li=Vi−Costi\text{P/L}_i = V_i - \text{Cost}_i

The portfolio percentage is computed on aggregated amounts over the set KK of positions with a known, non-zero cost, not as an average of per-line percentages, which would let small lines distort the result:

P/L%=∑i∈K(Vi−Costi)∑i∈KCosti\text{P/L\%} = \frac{\sum_{i \in K} \left(V_i - \text{Cost}_i\right)}{\sum_{i \in K} \text{Cost}_i}

When several lots of the same asset are merged, their total costs and values are summed before any ratio is taken. Positions without a known cost are excluded from both sides of the ratio and flagged, rather than assumed to have cost nothing (which would overstate your gain) or their current value (which would hide it).

Internal rate of return (IRR)

A simple percentage gain ignores when money went in and came out. The internal rate of return does not: it is the annual rate rr that makes the present value of all dated cash flows CFkCF_k equal to zero, using actual days elapsed (the convention of the spreadsheet function XIRR):

∑kCFk(1+r)(tk−t0)/365=0\sum_k \frac{CF_k}{(1 + r)^{(t_k - t_0)/365}} = 0

It is solved numerically.

Property

For real estate, the cash flows are:

  • the purchase price plus recorded acquisition costs, as an outflow on the purchase date;
  • each month, net rent received while a lease runs, minus running costs;
  • today's estimated value, treated as if the property were sold today.

Mortgage flows are deliberately left out, so the IRR measures the property's own performance, independent of how it was financed. Several properties can be combined into one portfolio IRR. The figure is withheld during roughly the first year of ownership, when annualising a short period mostly magnifies noise, and whenever the cash flows do not admit a meaningful solution.

Cash you have put in is estimated from the purchase costs and the loan principal repaid so far, since the app records loan balances rather than every payment.

Goal projections

A wealth goal projects today's value PVPV plus a level monthly contribution CC over nn months, compounding monthly at annual return rr, with i=r/12i = r/12:

FV=PV (1+i)n+C (1+i)n−1iFV = PV\,(1 + i)^n + C \, \frac{(1 + i)^n - 1}{i}

The required return is the rate r∗r^{*} that makes FVFV reach the target by the deadline. It has no closed form, so it is solved numerically by bisection, with no artificial ceiling. It is floored at 0% when contributions alone are enough.

The expected return is the value-weighted average of long-run return assumptions for the asset types the goal counts. Comparing the two gives the viability band: comfortable when r∗r^{*} is at or below the expected return, then moderate, difficult or unrealistic as the gap widens.

Monte Carlo simulation

A single projection assumes the future is average. The Monte Carlo planner instead simulates thousands of possible futures, year by year, and reports the spread of outcomes.

Inflation

Inflation is persistent: a high year tends to be followed by another. It follows a mean-reverting AR(1) process around your long-run assumption πˉ\bar{\pi}:

πt=πˉ+ϕπ(πt−1−πˉ)+σπ1−ϕπ2  εtπ\pi_t = \bar{\pi} + \phi_\pi \left(\pi_{t-1} - \bar{\pi}\right) + \sigma_\pi \sqrt{1 - \phi_\pi^2}\; \varepsilon^{\pi}_t

Asset returns

Each asset's annual shock combines a common part, correlated with inflation and with other assets of its group, and an idiosyncratic part drawn from a skew-normal distribution, which gives fat tails and asymmetric upside:

εa,t=β(ρa εtπ+1−ρa2  zg(a),t)+1−β2  ηa,t\varepsilon_{a,t} = \beta \left( \rho_a \, \varepsilon^{\pi}_t + \sqrt{1 - \rho_a^2}\; z_{g(a),t} \right) + \sqrt{1 - \beta^2}\; \eta_{a,t}

Group shocks zg,tz_{g,t} are correlated across four asset groups (growth, defensive, real assets, alternatives) through the Cholesky factor LL of the group correlation matrix, zt=L utz_t = L\,u_t with independent standard normals utu_t. Returns are log-normal and independent from one year to the next:

ℓa,t=μaτ+σaτ εa,t,Va,t=Va,t−1 eℓa,t\ell_{a,t} = \mu_a \tau + \sigma_a \sqrt{\tau}\, \varepsilon_{a,t}, \qquad V_{a,t} = V_{a,t-1}\, e^{\ell_{a,t}}

where μa\mu_a is the expected yearly price return (dividends are paid on top of it) and τ\tau pro-rates the current, partial year. Values can fall but never turn negative.

Cash flows

Each simulated year applies your plan in order:

  • income grows with the growth mode you chose (fixed, with inflation, and so on) and is taxed;
  • expenses are indexed to that path's own inflation;
  • loans amortise and dividends are paid on top of the price return;
  • a surplus is invested according to your savings rate;
  • a deficit is covered from savings first, then by selling assets in the priority you set, with capital-gains tax on the realised gain over each asset's cost basis.

A path is ruined in the first year a deficit can no longer be covered. Milestones (retirement, a purchase, a change of income) trigger on the conditions you define, evaluated on each path's own history.

What you see

  • Net worth bands: the 10th, 25th, 50th, 75th and 90th percentiles across all paths, for every year.
  • Success probability: the share of paths never ruined, and the ruin curve over time.
  • Milestone timing: the distribution of the year each milestone is reached.
  • A representative path: the median path of the most likely outcome, replayed in full so that its year-by-year income, expenses and assets are internally consistent.

Every run uses a seeded random generator, and paths are drawn in antithetic pairs (each draw mirrored by its opposite) to reduce sampling noise. The same plan and seed always reproduce the same result.

Limitations

Rigour includes stating limits plainly.

  • Not advice. Everything on Dollar Odyssey is information about your own portfolio. It is not investment, tax or legal advice.
  • Judgment, not prophecy. Country profiles, industry sensitivities and scenarios are structured, researched judgments. They describe tendencies, not forecasts or price targets.
  • Data lags. Fund holdings, company disclosures and country indicators are published periodically, so classifications reflect the latest available information, not today's exact composition.
  • Coverage varies. Smaller or newer securities may not be researched yet. The coverage line on each page tells you how much of your portfolio this affects.
  • Simulations depend on assumptions. Monte Carlo results are only as realistic as the return, volatility and inflation assumptions behind them. Treat the bands as a range of plausible outcomes, not a guarantee.

References

Standards and public data

Methods

  • Box, G. E. P. and Muller, M. E. (1958). A note on the generation of random normal deviates. The Annals of Mathematical Statistics, 29(2).
  • Azzalini, A. (1985). A class of distributions which includes the normal ones. Scandinavian Journal of Statistics, 12(2).
  • Hammersley, J. M. and Morton, K. W. (1956). A new Monte Carlo technique: antithetic variates. Mathematical Proceedings of the Cambridge Philosophical Society, 52(3).
  • Glasserman, P. (2003). Monte Carlo Methods in Financial Engineering. Springer.

READY FOR MORE

Start free, then step into deeper planning when the stakes get real

Create an account to keep your place in the workflow, then upgrade when you want richer simulations, gated calculators, and more powerful planning support.