API tools for institutional investors & hedge funds.
Bankster API
The tools behind institutional advantage.
Institutional expertise.
Working processes.
Competitive intent.
BANKSTER API / BLOG
Signed opinion from the desk and the back office: positions on data, money, controls and the systems institutional finance runs on.
Vendor and exchange terms distinguish display from non-display use, restrict derived data and redistribution, and price by user, application, or location. Moving analytics to the cloud, adding a data platform, or feeding an AI model can change what you owe.
Read postPCI DSS compliance is a commercial requirement to keep processing cards, not a security warranty. The attestation is a point-in-time snapshot of a sampled environment, and the real work is scope reduction and the risk narrative that sits beside it.
Read postRescanning the full trade tape for every VWAP report is a costly way to avoid addition. Store running sums of price times quantity and quantity per bucket, and divide once when reading.
Read postRegulators increasingly expect every reported number to be traceable to its source and reproducible on demand. Most reports are assembled downstream, through extracts, adjustments, and spreadsheets that nobody can fully trace. Lineage has to be designed into the data flow; it cannot be reconstructed the week before a regulatory review.
Read postBinary floating-point arithmetic is not associative. Change the summation order, the thread count, or the hardware, and the risk number changes in the last digits—sometimes more. Where the differences matter, where they don't, and how to set tolerances that can be defended.
Read postA practitioner's guide to choosing between DynamoDB and PostgreSQL for institutional financial systems, based on query volatility, invariants, precision, and economics.
Read postBinary floating point can't represent 0.1 exactly. The screen rounds it politely; the reconciliation team doesn't. Money needs an exact representation and a rounding rule decided in advance.
Read postLanguage models compress away the hedged language and conditional scenarios that risk committees most need to see.
Read postExcel sees ISINs and CUSIPs that look like dates or numbers and helpfully converts them. Identifiers are labels, not numbers, and a spreadsheet that rewrites them is inventing reference data.
Read postA row-oriented transactional database will not answer that in seconds, however many indexes you add. Matching the analytical architecture to how analysts actually query.
Read postIf positions arrive overnight, the dashboard isn't real-time; it's punctual. A refresh timestamp says when the screen was rebuilt, not how old the numbers are.
Read postSOC 2 is an auditor's attestation, not a certification, and a Type I report says less than a Type II. Whether you need one depends on who your clients are, what they would otherwise send you, and whether ISO 27001 or a client's own audit rights would serve them better.
Read postA new instrument type is not live when the order management system can trade it; it is live when the month-end close succeeds.
Read postWhat to check before retrofitting an office building for core financial infrastructure, and when colocation is the better answer.
Read postWhen a regulated financial workload moves from a relational database to a document or key-value store, the controls do not disappear. They migrate into application code, where they are harder to audit, harder to test, and easier to bypass.
Read postSome figures do not aggregate. The average of fund-level internal rates of return is not the portfolio's IRR; that requires pooling the underlying cash flows. Money-weighted and time-weighted returns answer different questions; percentages and ratios summed or averaged across rows produce numbers that look precise and mean nothing.
Read postJPMorgan's 2012 trading loss was first disclosed at around $2 billion and grew to roughly $6.2 billion. The bank's own review found that the value-at-risk model behind that trading ran on spreadsheets fed by manual copy and paste, with a formula that divided by a sum where it should have divided by an average, understating the risk. What the case says about end-user computing in risk management, model change control, and who reviews the arithmetic.
Read postAverage a pauper's 100 percent gain with a king's zero and you get 50 percent growth for a kingdom that grew by a dollar. Averaging percentages answers a different question than the one you asked.
Read postThe LIBOR transition was less a rate change than a forced inventory of every system, contract, and spreadsheet that touched a number nobody could trace.
Read postWhat a 2008 trading loss reveals about access reviews, segregation of duties, and the alerts nobody follows up.
Read postWhy institutional platforms built for daily prices and standardized feeds break down when they encounter capital calls, quarterly marks, and PDFs from general partners.
Read postWhy rounding rules belong in requirements, data contracts, and migration test plans.
Read postReplacing a pension administration platform is mostly a data-quality and knowledge-recovery project: missing dates, rules that exist only in code, and calculations nobody can reproduce by hand.
Read postCustodians hold the assets, but over time they also become the data platform, the reporting engine, the accounting provider, and the source of dozens of integrations built on their message formats. Switching becomes a multi-year program nobody wants to sponsor.
Read postAn ISO 27001 certificate proves a vendor runs a risk management process, not that its services are secure. The scope statement and Statement of Applicability are where the real answer lives.
Read postThe richer data in ISO 20022 only helps if every system in the payment chain can carry it. Truncation at the narrowest database column turns a richer standard into a reconciliation and compliance problem.
Read postEvery trade, position, price, and report depends on reference data. The security master is rarely anyone's flagship project, and a wrong record in it propagates everywhere.
Read postA shared ledger can prove both firms recorded the same message. It cannot make them record the same trade. Reconciliation breaks start upstream, and that is where they get fixed.
Read postAI changes the economics of compliance reading, but restructuring around it is a decision-rights exercise with a headcount consequence.
Read postPolicies written by pasting every clause of a framework and hardening the verbs look complete and fail on first contact with an auditor. Auditors test whether controls operate and whether evidence exists, and standards like ISO 27001 expect controls chosen from a risk assessment, not adopted wholesale. A policy you do not follow is documented non-compliance. Writing policies the organization can actually run.
Read postA global portfolio has no single close. Which price counts for a Tokyo holding in a Toronto fund, when the NAV cut-off falls, and when fair-value adjustments apply are business decisions that end up hard-coded differently in each system.
Read postWhat T+1 settlement does to the overnight batch, the affirmation chain, FX funding, and securities lending recalls—and why faster hardware will not fix a sequential dependency problem.
Read postDatabase schemas encode assumptions about precision. Fractional shares, high-precision FX rates, and tokenized assets break those assumptions in production.
Read postModel risk frameworks written for pricing engines now reach valuation spreadsheets, end-user computing tools, vendor analytics, and machine learning models. Inventory, validation, and ownership for models nobody on the IT side knew existed.
Read postA decision guide for regulated institutions evaluating a move off Microsoft 365, covering archives, legal holds, add-ins, and the dependencies that decide the outcome.
Read postThe investment book, the accounting book, and the custodian each hold a different version of the same position. The operating model only has to prove it knows exactly why.
Read postTrading systems with microsecond latency justify exotic engineering. A weekly client-newsletter chart doesn't. It isn't failing for want of a type system; it's failing for want of an owner.
Read postIf the morning risk numbers depend on one person's laptop, the laptop is critical infrastructure. A notebook can calculate risk; operating a risk process takes ownership, reproducibility and recovery.
Read post