MDO · Financial Intelligence Engine

Your Data.
Your Factors.
One Day.

One analyst. One day of MDO Surge prompts. Built the factors, wired the pipelines, ready to test any frequency, any history, any universe. In this case: 50 factors, 25 years of history, 3,000 global stocks, point-in-time accurate, live in your Snowflake. AI generates answers. MDO ensures they’re correct.

50
Production-grade factors built in one day — a starting point, not a ceiling
25Y
History · Jan 2000 – present · monthly rebalance
3,000
Global stocks · top market cap · survivorship-free
912K
Rows in your Snowflake · PIT-aligned · FX-normalized
Before MDO · After MDO
Typical in-house build: a project. Engineering, planning, vendor integrations — the factory before the factors.
With MDO: skip the factory. Use the engine.
Worldscope · IBES Estimates · Datastream  |  Built in one day with MDO Surge + the Financial Intelligence Engine
The Engineering Tax

Six problems your team solves before testing a single factor.

None of these are factor research. All of them must be solved first. MDO ships every one of them built in.

Problem Why it’s hard How MDO handles it
Point-in-time alignment Raw warehouse data is restated — look-ahead bias by default. ✓  dateAlign is a built-in parameter on every fundamental call — prelim, report-date, fiscal-period-end, and other PIT conventions are supported out of the box. Look-ahead bias is never an afterthought.
Survivorship-bias-free universe Delisted and bankrupt names disappear from current constituent lists. ✓  QAD-sourced monthly constituent history rebuilds the top-3,000 universe at each month-end.
Cross-currency normalization Global factors need FX as-of the calculation date, per security, every call. ✓  toCurrency converts to USD, EUR, JPY, GBP, or any supported currency inline — FX rates applied as-of the factor date. No separate FX pipeline, regardless of base currency.
TTM across reporting frequencies US quarterly, Japan semi-annual, EU annual — one calc must handle all three. ✓  Batched Q/S/A pulls with ttm=True and Q→S→A coalesce. No bespoke logic.
Reporting lag enforcement A Q3 filing published in Feb can’t be usable in Oct without dedicated lag logic. ✓  Preliminary filing lag enforced per company. restated=False locks the originally reported figure.
Snowflake-native execution Pandas-based factor logic forces data movement and breaks at scale. ✓  Factors run as Snowpark UDFs. Data never leaves your account. Scales with your warehouse tier.
Pipeline Architecture

From raw data to 912,000 factor rows — one pipeline, your infrastructure.

Three LSEG sources through MDO’s Financial Intelligence Engine, into a point-in-time accurate factor dataset in your Snowflake. Every number below is real — measured end-to-end on a standard Snowflake warehouse.

LSEG DATA SOURCES MDO FINANCIAL INTELLIGENCE ENGINE MDO SURGE FACTOR LIBRARY Worldscope Fundamentals · PIT-aligned Balance sheet · Income · Cash flow IBES Estimates Consensus EPS · Revenue Median estimates · Revision counts Datastream Pricing · Returns · FX DS_CLOSE · DS_TOTRET · DS_MKTCAP Financial Intelligence Engine DateAlign · TTM · FX · PIT Survivorship-free universe Snowflake-native UDFs Reporting lag management Factor staging + caching Universe: 3,000 stocks × 304 months 25-year history · 2000 → today your Snowflake account MDO Surge Plain-language prompt → production Python code Financially correct item codes Validated against FIE catalog Analyst refines, not engineers Factor Library 50 built in one day Value × 10 Quality × 10 Growth × 10 Momentum × 8 Risk × 7 Technical × 5 912,000 rows PIT · FX-USD · plain Python Your Investment Team Backtesting · Portfolio construction · Signal research · New hypotheses Your data. Your infrastructure. Your 50 factors — or 500. MDO's job is done. ⚡ Built end-to-end in one day

Measured compute times — 25-year full backfill, 3,000 securities

Universe build (top 3,000 global)
139 seconds · 25Y full backfill
FL_MOM_SURP (IBES surprise, fastest)
40 seconds · 25Y full backfill
FL_VAL_EVEBITDA (EV/EBITDA, median)
170 seconds · 25Y full backfill
FL_RSK_BETA (12M OLS regression, slowest)
234 seconds · 25Y full backfill
50
factors loaded
into your Snowflake
91 min
factor compute
Snowpark UDFs · no pandas
~8 min
incremental daily update
just the latest month

Your cloud. Your control.

Every factor table lives in your Snowflake account. Data never moves, never replicates, never egresses. Compute scales with the warehouse tier you already pay for — compliance-friendly by construction, no new vendor surface area to audit.

Meet Surge · MDO's AI Agent

Two paths to the same answer. Both correct by construction.

AI generates answers. MDO ensures they’re correct. Same engine, two interfaces:

For quants & engineers
MDO Surge → editable Python
Prompt in plain English, get production Python code you can audit, modify, and extend.
For PMs & analysts
Claude via MCP → no code at all
Ask investment questions directly inside Claude. MDO returns institutional-grade answers.
Analyst Prompt to MDO Surge
"Price acceleration — is the stock moving faster in the last 3 months versus the 3 months before that?"
↓   Surge generates production code
FL_TEC_PRICEACC · Price Acceleration
def FL_TEC_PRICEACC(universe, params=None):
    import mdosnow as mdo
    from snowflake.snowpark.functions import expr
    dims = ['day', 'security']
    universe = mdo.as_universe(universe[dims], dimensions=dims)

    # One call fetches both 3M and 6M returns simultaneously
    mdo.PlugDataChange(universe, items={'RET': 'DS_TOTRET'},
                       shift=[-3, -6], periodType='month',
                       toCurrency='USD', recent=-10, units='P')

    # Acceleration = current 3M minus prior 3M (non-overlapping)
    # Prior 3M ≈ 6M - current 3M  =>  2×3M - 6M
    universe.with_column('FL_TEC_PRICEACC',
                         expr('2.0 * RET_T3_CHG - RET_T6_CHG'))
    return mdo.as_universe(universe[dims + ['FL_TEC_PRICEACC']], dimensions=dims)
Analyst Prompt to MDO Surge
"Return on equity — TTM net income over average shareholder equity. Make sure semi-annual reporters and companies with only one year of equity history are still included."
↓   Surge generates production code
FL_QL_ROE · Return on Equity
def FL_QL_ROE(universe, params=None):
    import mdosnow as mdo
    from snowflake.snowpark.functions import coalesce, expr
    dims = ['day', 'security']
    ws = dict(dateAlign='prelim', restated=False, toCurrency='USD', units='M')
    universe = mdo.as_universe(universe[dims], dimensions=dims)

    # TTM net income — Q/S/A in one call, coalesced for global coverage
    # (US quarterly · Japan/Korea semi-annual · EU annual all handled)
    mdo.FundamentalData(universe, {
        'NI_Q': 'WS_1651_Q', 'NI_S': 'WS_1651_S', 'NI_A': 'WS_1651_A',
    }, period=0, ttm=True, **ws)
    universe.with_column('NI_TTM', coalesce('NI_Q', 'NI_S', 'NI_A'))

    # Current book equity — Q/S/A coalesced to most-recent reported value
    mdo.FundamentalData(universe, {
        'EQ_Q': 'WS_3995_Q', 'EQ_S': 'WS_3995_S', 'EQ_A': 'WS_3995_A',
    }, period=0, **ws)
    universe.with_column('EQ', coalesce('EQ_Q', 'EQ_S', 'EQ_A'))

    # Prior-year equity (annual at period=-1; MDO writes column as EQ_T1).
    # If prior is missing, fall back to current equity so the row isn't dropped —
    # critical for recent IPOs and stub-year filings.
    mdo.FundamentalData(universe, items={'EQ': 'WS_3995_A'}, period=-1, **ws)

    universe.with_column('FL_QL_ROE',
        expr('NI_TTM / NULLIFZERO((EQ + COALESCE(EQ_T1, EQ)) / 2.0)'))
    return mdo.as_universe(universe[dims + ['FL_QL_ROE']], dimensions=dims)
Analyst Prompt to MDO Surge
"Earnings surprise — how much did the company beat or miss analyst consensus last quarter?"
↓   Surge generates production code
FL_MOM_SURP · Earnings Surprise
def FL_MOM_SURP(universe, params=None):
    import mdosnow as mdo
    from snowflake.snowpark.functions import coalesce
    dims = ['day', 'security']
    universe = mdo.as_universe(universe[dims], dimensions=dims)

    # IBES pre-computed standardized surprise: (Actual - Consensus) / |Consensus|
    # Q/S/A in one call — US quarterly, Japan/Korea/Australia semi-annual,
    # EU annual all coalesced into one signal. dateAlign='prelim' enforces PIT.
    mdo.FundamentalData(universe, {
        'SURP_Q': 'IB_ACT_EPS_SURPRISE_Q',
        'SURP_S': 'IB_ACT_EPS_SURPRISE_S',
        'SURP_A': 'IB_ACT_EPS_SURPRISE_A',
    }, period=0, dateAlign='prelim', restated=False)

    universe.with_column('FL_MOM_SURP', coalesce('SURP_Q', 'SURP_S', 'SURP_A'))
    return mdo.as_universe(universe[dims + ['FL_MOM_SURP']], dimensions=dims)
Not just an AI chatbot on top of raw data. Surge knows the correct LSEG item codes, needs NULLIFZERO in ratio formulas, and handles Japanese semi-annual reporters in global TTM calcs. The financial domain knowledge lives in the Financial Intelligence Engine — not the prompt — so every output is correct, reproducible, and auditable.
New · MDO + Claude via MCP

Investment research without code — now inside Claude.

MDO is now an MCP server integrated with Claude. Ask in plain English — factor research, screening, backtesting — and Claude runs it against MDO’s hardened data layer. The responsibility shifts to asking the right question.

“Run a Russell 3000 value-momentum backtest, last 10 years, monthly rebalance.”
→ Claude calls MDO via MCP
→ Point-in-time accurate, FX-normalized
→ Pre-built, audited workflow
→ Reproducible Python in your Snowflake
Factor Library

18 of the 50. Every formula is plain, editable Python.

A curated cross-section — three per category — from a library built in one day. Add, modify, or extend any factor: the framework scales with your research.

Factor Signal Category Source Formula Coverage
FL_VAL_EP Earnings Yield Value Worldscope WS_1651_Q (TTM NI) / DS_MKTCAP
98%
FL_VAL_BP Book-to-Price Value Worldscope WS_3995_A / DS_MKTCAP
98%
FL_VAL_DIVYIELD Dividend Yield Value Datastream DS_DY (direct)
98%
FL_QL_ROE Return on Equity Quality Worldscope TTM NI / Avg Equity
98%
FL_QL_ROA Return on Assets Quality Worldscope TTM NI / Avg Total Assets
98%
FL_QL_OPMGN Operating Margin Quality Worldscope TTM EBIT / TTM Revenue
96%
FL_GR_EPSREV EPS Revision Score Growth IBES (Up30D − Down30D) / NumEst
94%
FL_GR_SALESGROW Sales Growth 1Y Growth Worldscope TTM Revenue YoY
94%
FL_GR_EPSGROW1Y EPS Growth 1Y Growth Worldscope TTM NI YoY / |Prior|
94%
FL_MOM_12M1M 12M1M Price Momentum Momentum Datastream 12M Return − 1M (skip-month)
97%
FL_MOM_SURP Earnings Surprise Momentum IBES IB_ACT_EPS_SURPRISE_A
92%
FL_MOM_52WHIGH 52-Week High Proximity Momentum Datastream Price / Rolling 52W High
100%
FL_RSK_BETA Market Beta Risk Datastream OLS β vs EW universe (12M)
97%
FL_RSK_IDIOVOL Idiosyncratic Volatility Risk Datastream StdDev of Market-Adj Residuals
97%
FL_RSK_VOLRET Return Volatility Risk Datastream 12M StdDev of Monthly Returns
100%
FL_TEC_52WHIGH Annual Range Position Technical Datastream (Price − 52W Low) / (High − Low)
100%
FL_TEC_PRICEACC Price Acceleration Technical Datastream 2 × 3M Return − 6M Return
98%
FL_TEC_TURN Share Turnover Technical Datastream 3M Avg Volume / Implied Shares
100%

A note on coverage: percentages reflect what global companies actually report — not MDO limitations. Lower-coverage factors apply where the underlying data exists; the survivorship-safe universe is preserved otherwise. Full 50-factor library available on request.

Custom Formulas

Plain Python, registered as a managed MDO Custom Formula.

Every factor isn’t just code — it’s a Custom Formula in your Snowflake. Once registered, anyone in your user group can query it like any native MDO data item, with full PIT, FX, and survivorship guarantees baked in.

1
Write plain Python
def FL_VAL_MYMETRIC(universe, params=None):
Analyst writes the factor as a function. Editable, auditable, in source control.
→
2
Register as Custom Formula
mdo.CFCreate(name='FL_VAL_MYMETRIC', formula=FL_VAL_MYMETRIC, ...)
MDO registers it as a managed Custom Formula in your formula group.
→
3
Load into your Snowflake
mdo.CFDataLoad(universe, name='FL_VAL_MYMETRIC', ...)
Values computed and persisted to a managed table in your account.
→
4
Anyone queries it
mdo.CFData(universe, name='FL_VAL_MYMETRIC', ...)
Other analysts, models, and dashboards consume it like any MDO item.
This is not a folder of scripts. It’s a versioned, managed catalog of factors living inside your Snowflake. Edit the Python → re-register the Custom Formula → reload — downstream consumers keep working without code changes. The factor library becomes institutional infrastructure, not a side project.

From there it’s your workflow: connect the factor tables (or the consolidated CSV) to your existing backtesting, portfolio-construction, and risk-attribution tools. Your data, your tools, your edge.

Why MDO

AI generates answers. MDO ensures they're correct.

Generic AI on raw data writes plausible code. MDO’s investment-ready framework makes the answer institutional-grade: point-in-time accurate, survivorship-safe, FX-normalized, reproducible.

Capability Generic AI on Your Data MDO Surge + Financial Intelligence Engine
Financial domain knowledge ◦  Generates syntactically valid code — financial correctness depends on the prompt ✓  Generates code with correct LSEG item codes validated against the FIE catalog
Point-in-time alignment ◦  Has no concept of when data was available — look-ahead bias is possible ✓  DateAlign is a built-in parameter on every data call — PIT is never an afterthought
Global reporting frequencies ◦  Treats all companies as quarterly reporters by default ✓  TTM logic handles quarterly, semi-annual, and annual fiscal calendars automatically
FX normalization ◦  FX conversion is left to your code — rates, timing, and source are your problem ✓  toCurrency='USD' on any data call — rates sourced from Datastream as-of the factor date
Survivorship bias ◦  Generated code queries your current universe — delisted stocks require separate handling ✓  QAD-sourced universe includes all delisted tickers through their last active date
Execution environment ◦  Generates pandas code — runs locally or requires ETL into your warehouse ✓  Pushes execution into Snowflake UDFs — data never leaves your cloud
Adding new data items ◦  Model may hallucinate item codes; validation requires manual lookup ✓  Item codes are catalogued in the FIE; Surge references only validated codes
The intelligence isn’t just the language model. It’s the Financial Intelligence Engine underneath — point-in-time alignment, the survivorship-safe universe, the validated item catalog, the Snowflake-native execution layer. Across over 70 data sources, any time series, any universe.
Your Next Step

Ready to build your factor library?

MDO connects directly to over 70 data sources inside Snowflake — point-in-time accurate, FX-normalized, institutional-grade. Now available with Surge and Claude via MCP. No data movement, no replication lag, no infrastructure to manage.

Talk to MDO →