What is point-in-time financial data?

Point-in-time data preserves both the period a fact describes and the date that fact became knowable. A historical query returns only the filings, values and revisions available by the chosen date rather than today’s cleaned-up history.

Why is point-in-time data important for backtesting?

Without it, a backtest can use restated accounts, later corporate-action adjustments or revised classifications before those changes were public. That lookahead bias can make a strategy appear stronger than it could have been in real time.

How should an investment team test a financial-data provider?

Ask the provider to reconstruct one company as of a past date, show original and revised values side by side, explain consolidated-versus-standalone selection, demonstrate corporate-action handling and reproduce a derived metric from its dated inputs.

Should missing financial data be treated as zero?

No. Missing, not applicable and genuinely zero are different states. Converting absence into zero creates false signals, distorts ratios and hides coverage problems.

Point-in-Time Financial Data in India: A Buyer’s Guide for Investment Teams

A point-in-time financial database should be able to recreate what an investment team could actually have known on a chosen date. If a provider can show only today’s restated history, it may be useful for current analysis but unsafe for historical screens, backtests and decision audits.

The most important buying question is therefore not “how many years of data do you have?” It is:

Can you reproduce the exact information set available to an analyst before a historical decision, including later revisions that did not yet exist?

This guide turns that principle into a practical evaluation for PMS firms, AIFs, family offices, research desks and quantitative teams buying Indian financial data.

First, separate three different dates

Many data products attach one date to a financial value. Serious point-in-time research needs at least three concepts.

The period end tells you what economic period the number describes. Revenue for the quarter ended 30 June belongs to that quarter.
The publication or availability date tells you when the number became knowable. The June-quarter revenue may be published in July or August.
The observation or ingestion date tells you when the data system captured that version. This is operational metadata, not a substitute for the public availability date.

Confusing period end with publication date is the simplest form of lookahead bias. A March financial year does not make the annual result knowable on 31 March.

Ask the provider to show these fields explicitly. A generic “date” column is not enough.

Test 1: Can the system reconstruct a historical view?

Choose a company that later restated a prior period, changed segment reporting, completed a demerger or classified a business as discontinued.

Then ask for two outputs:

the company history as it was knowable immediately before the later change;
the latest restated history available today.

Both series can be correct. They answer different questions.

The first is appropriate for replaying a decision or backtesting a rule. The second may be more comparable for analysing the company’s current economic history. A mature provider preserves both instead of silently overwriting the earlier record.

If the two requests return the same values without an explanation, the database may be versioned by period but not genuinely point-in-time.

For a plain-English explanation of the underlying risk, read why point-in-time data matters.

Test 2: Ask to see the vintages

A vintage is a version of a fact as it existed at a particular observation point. The original filing, a company correction and a later restatement should not collapse into one mutable row.

For one financial line item, ask the vendor to show:

Field	What you should expect
Company and statement basis	The legal entity and consolidated or standalone basis
Fiscal period	Quarter or year the fact describes
Value and unit	The reported amount with an unambiguous unit
Source	Filing or primary document from which it came
Available from	When the market could first know this version
Vintage seen at	When the system captured the version
Revision relationship	Whether it supersedes, corrects or restates an earlier fact

The provider does not need to use these exact column names. It does need to preserve the concepts.

Test 3: Check consolidated and standalone discipline

Indian listed companies frequently publish both consolidated and standalone accounts. The two bases are not interchangeable.

A clean time series should generally prefer consolidated statements when they represent the group, fall back to standalone only under a defined rule and never mix bases inside a calculation window without making that switch visible.

Ask the provider:

Which basis wins when both are available?
What happens when consolidated results start only midway through history?
Can a user see which basis supports every period?
Can a four-quarter calculation accidentally combine three consolidated quarters with one standalone quarter?

A chart can look smooth while its accounting basis changes underneath it. Metadata matters as much as presentation.

Test 4: Examine corporate-action handling

Prices, shares and per-share metrics require a consistent corporate-action basis. Splits, bonuses, rights issues, demergers and dividends can make a historical series misleading if the price is adjusted but the share count or earnings series is not.

Choose a company with a known split or demerger. Ask the provider to reproduce:

the raw historical price;
the adjusted price and adjustment factor;
the share-count timeline;
earnings per share on the matching basis;
the effective date and source of the action.

A low historical P/E created by pairing an adjusted price with unadjusted earnings is not an investment insight. It is a basis error.

Test 5: Reproduce one derived metric

Point-in-time discipline must extend above raw statements.

Suppose a provider supplies ROCE, earnings growth, valuation ratios or a forensic score. Ask it to calculate one historical observation from the exact dated inputs available at that point.

The provider should be able to explain:

the formula and sector applicability;
the statement basis;
the periods included in the numerator and denominator;
the availability date of every input;
the rule used when an input is unavailable;
the effective availability date of the final metric.

For a composite metric, the output cannot become knowable before its latest required input. That simple maximum-of-inputs rule prevents a surprising amount of accidental hindsight.

Test 6: Make absence visible

Missing is not zero.

A company with no applicable inventory figure is different from a company reporting zero inventory. An unavailable dividend record is different from a confirmed zero dividend. A ratio that cannot be calculated because an input is absent is different from a ratio of zero.

Ask the provider to distinguish at least:

reported zero;
not reported;
not applicable;
not yet available;
outside coverage;
failed validation.

If every gap becomes 0, screens will select data defects as if they were investment signals. A good system exposes the missing state and, ideally, the reason.

Test 7: Demand a reproducible historical query

Ask the provider to write down, in ordinary language, the answer to this request:

Return the latest annual and trailing financial information that was publicly available for every eligible company at market close on 31 March 2021.

Then rerun the same request twice.

The universe, values, eligibility logic and timestamps should be reproducible. If results change, the provider should be able to identify the new vintage or code version that caused the change.

This is essential for research governance. A strategy result without a data snapshot, universe definition and calculation version is difficult to audit and nearly impossible to learn from.

Test 8: Inspect universe and survivorship policy

Point-in-time values do not rescue a backtest built on today’s survivors.

Ask whether the historical universe includes companies that were later delisted, merged, renamed, suspended or removed from an index. Ask how listing dates and security-identity changes are handled. Ask whether current index membership is being projected backward.

The provider should also explain quality gates. A company can exist in the historical universe while lacking enough valid, timely data for a particular rule. Excluding it silently can create another bias; treating the missing values as zero is worse.

A practical RFP checklist

Before buying point-in-time Indian financial data, ask for evidence against each item.

Do not accept a slide that says “PIT-ready” in place of a live reconstruction. Give the vendor a difficult company and a difficult date.

Where Altys fits

Altys point-in-time financial data is designed around the difference between the period a fact describes, the date it became knowable and the later vintages that may revise it.

The same discipline is intended to flow into derived metrics, screens, factor studies and strategy backtests. Source links, statement basis, availability rules and explicit unavailable states are not supporting details. They determine whether the resulting research can be defended.

Altys also connects that historical data layer with company research and monitoring. A team can test an investment rule using the information available then, record the thesis formed from that evidence and later compare the expectation with the outcome. This closes the loop between backtesting, live research and post-decision learning.

The buying principle

The cleanest financial history is not always the history an investor saw.

Buyers should therefore evaluate a data product on two separate abilities:

Can it present the best current understanding of a company’s history?
Can it reconstruct the imperfect, partial information set that existed at a past decision date?

Serious research needs both. The second is what makes the data genuinely point-in-time.