Morgan Stanley Data Analyst Interview Guide

Most candidates prepping for Morgan Stanley load up on Python and machine learning. That's a miscalibration. From hundreds of mock interviews we've run for bank data roles, the people who struggle here can write a gradient boosting pipeline but can't explain what a suspicious wire transfer pattern looks like or design a normalized schema for advisor performance tracking.

Morgan Stanley Data Analyst Role

This role lives at the intersection of Wealth Management reporting and the firm's compliance infrastructure. You'll spend most of your time in Hive, Impala, and Tableau, pulling from client relationship tables and transaction schemas to answer questions from financial advisors, regional directors, and risk officers. Success after year one means you own recurring reporting packages (the weekly AUM/flows deck, the FA productivity scorecard in Tableau) and stakeholders trust your numbers enough to present them to leadership without re-checking.

A Typical Week

What the time split won't tell you is how reactive the work feels. A regional director in the Southeast pings you about a client attrition spike, and suddenly your afternoon plan to automate the eVestment benchmarking pull is gone. The other thing that catches new hires off guard: debugging broken data feeds (a Morningstar identifier format change blowing up a LEFT JOIN in your Hive staging tables) is a recurring part of the job, not a rare fire drill.

Projects & Impact Areas

Fraud investigation and anomaly detection feed directly into compliance workflows, which makes them high-visibility for the role. On the Wealth Management side, you're building advisor performance metrics and client-facing reporting that leadership reviews weekly. A less obvious but growing pocket sits within Parametric, Morgan Stanley's systematic investment arm, where data analysts support investment proposal pipelines and data integration work touching eVestment and fund performance data.

Skills & What's Expected

SQL will make or break your candidacy. Machine learning scores medium in importance for this role, meaning basic familiarity is expected, but interviewers spend far more time on relational querying, statistics, and finance domain knowledge. What's genuinely underrated: data visualization and communication. You'll present findings to non-technical stakeholders on a weekly cadence using Tableau and Excel, and the ability to write a clean methodology doc matters more here than knowing scikit-learn's API.

Levels & Career Growth

Most external hires land at Analyst or Associate, with the dividing line around four years of experience. What separates the two isn't technical depth alone; it's whether you can independently scope an analysis, manage stakeholder expectations, and document your work to audit standards. The promotion blocker from Associate to VP is almost always influence without authority, your ability to drive decisions across teams (risk, compliance, data engineering) you don't control.

Work Culture

Morgan Stanley's hybrid policy requires at least three days in the New York office, but most Wealth Management analytics teams default to four because ad-hoc stakeholder requests hit constantly. Expect an 8:30 AM to 6:30 PM rhythm, with intensity spiking around month-end and quarter-end reporting cycles. The benefits are genuinely strong (backup childcare, tuition reimbursement, solid 401(k) match), but if "auditability and documentation" sounds tedious rather than reassuring, this probably isn't the right seat for you.

Morgan Stanley Data Analyst Compensation

Morgan Stanley's comp structure pairs a base salary with a discretionary annual bonus tied to both your individual rating and firm-wide results. RSUs might enter the picture at more senior levels, vesting over three to four years, but the source of truth on whether you'll see equity is your specific offer letter. The annual bonus is discretionary, not guaranteed, so don't mentally bank on a particular number when evaluating your total package.

Base salary and sign-on bonus are the most negotiable levers at offer stage. If you're holding a competing offer from JPMorgan or Goldman Sachs, the sign-on is where recruiters have real flexibility. Push there rather than grinding on base, which tends to move in smaller increments. One mistake candidates make: confusing Morgan Stanley investment banking compensation with what a Data Analyst offer looks like, then anchoring expectations to the wrong number entirely.

Morgan Stanley Data Analyst Interview Process

Budget about seven weeks from first contact to offer. The early Technical Assessment isn't just SQL. It covers statistics, algorithms, and data structures too, so candidates who treat it as a pure querying exercise walk in underprepared. Then Super Day hits you with a second, deeper SQL & Data Modeling round plus a live case study and behavioral, all back to back.

The most common rejection drivers, from what candidate reports suggest, are a cluster: shallow technical skills, limited finance understanding, and poor problem-solving approach during the case study. Acing the SQL portions won't save you if you can't reason through a fraud scenario or explain why you want to work at a financial services firm rather than a tech company. Prepare across all three fronts, because strength in one area doesn't appear to compensate for weakness in another.

Morgan Stanley Data Analyst Interview Questions

1WITH approved_cp AS (
2  SELECT
3    t.txn_id,
4    t.txn_ts,
5    CAST(t.txn_ts AS DATE) AS txn_dt,
6    t.amount
7  FROM transactions t
8  WHERE t.status = 'APPROVED'
9    AND t.card_id IS NOT NULL
10),
11cb_30d AS (
12  -- One row per transaction if it has any chargeback within 30 days of the transaction time
13  SELECT
14    a.txn_id,
15    1 AS has_cb_30d
16  FROM approved_cp a
17  JOIN chargebacks c
18    ON c.txn_id = a.txn_id
19   AND c.cb_ts >= a.txn_ts
20   AND c.cb_ts < a.txn_ts + INTERVAL '30' DAY
21  GROUP BY a.txn_id
22)
23SELECT
24  a.txn_dt,
25  SUM(a.amount) AS approved_dollars,
26  SUM(CASE WHEN cb.has_cb_30d = 1 THEN a.amount ELSE 0 END) AS chargebacked_dollars_30d,
27  CASE
28    WHEN SUM(a.amount) = 0 THEN NULL
29    ELSE SUM(CASE WHEN cb.has_cb_30d = 1 THEN a.amount ELSE 0 END) * 1.0 / SUM(a.amount)
30  END AS fraud_rate_30d
31FROM approved_cp a
32LEFT JOIN cb_30d cb
33  ON cb.txn_id = a.txn_id
34GROUP BY a.txn_dt
35ORDER BY a.txn_dt;

1WITH logins AS (
2  SELECT
3    e.account_id,
4    e.event_ts,
5    e.result,
6    CASE WHEN e.result = 'SUCCESS' THEN 1 ELSE 0 END AS is_success,
7    CASE WHEN e.result = 'FAIL' THEN 1 ELSE 0 END AS is_fail
8  FROM auth_events e
9  WHERE e.event_type = 'LOGIN'
10),
11segmented AS (
12  SELECT
13    l.*,
14    -- Each SUCCESS increments the segment id, so FAILs share the same segment until the next SUCCESS
15    SUM(l.is_success) OVER (
16      PARTITION BY l.account_id
17      ORDER BY l.event_ts
18      ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
19    ) AS success_segment
20  FROM logins l
21),
22fail_streaks AS (
23  SELECT
24    s.account_id,
25    s.success_segment,
26    COUNT(*) AS fail_streak_len
27  FROM segmented s
28  WHERE s.is_fail = 1
29  GROUP BY s.account_id, s.success_segment
30),
31account_max AS (
32  SELECT
33    fs.account_id,
34    MAX(fs.fail_streak_len) AS max_consecutive_failures
35  FROM fail_streaks fs
36  GROUP BY fs.account_id
37)
38SELECT
39  am.account_id,
40  am.max_consecutive_failures
41FROM account_max am
42ORDER BY am.max_consecutive_failures DESC, am.account_id
43FETCH FIRST 20 ROWS ONLY;

1from __future__ import annotations
2
3import pandas as pd
4import numpy as np
5
6
7def daily_decline_metrics(auth_events: pd.DataFrame) -> pd.DataFrame:
8    """Compute daily decline metrics for card authorization events.
9
10    Expected columns:
11      - account_id
12      - event_ts: timezone-aware UTC timestamps preferred
13      - amount
14      - decision: 'APPROVE' or 'DECLINE'
15      - decline_reason: nullable
16
17    Returns columns:
18      - date (YYYY-MM-DD, as datetime64[ns])
19      - total_auths
20      - declines
21      - decline_rate
22      - pct_declines_suspected_fraud
23    """
24
25    df = auth_events.copy()
26
27    # Robust timestamp parsing
28    df["event_ts"] = pd.to_datetime(df["event_ts"], utc=True, errors="coerce")
29    # Drop rows with invalid timestamps, cannot be assigned to a day
30    df = df.dropna(subset=["event_ts"])
31
32    df["date"] = df["event_ts"].dt.floor("D").dt.tz_convert("UTC").dt.tz_localize(None)
33
34    df["is_decline"] = df["decision"].eq("DECLINE")
35    df["is_suspected_fraud_decline"] = df["is_decline"] & df["decline_reason"].fillna("").eq("FRAUD_SUSPECTED")
36
37    agg = (
38        df.groupby("date", as_index=False)
39        .agg(
40            total_auths=("decision", "size"),
41            declines=("is_decline", "sum"),
42            suspected_fraud_declines=("is_suspected_fraud_decline", "sum"),
43        )
44    )
45
46    # Safe rates
47    agg["decline_rate"] = np.where(
48        agg["total_auths"].to_numpy() == 0,
49        0.0,
50        agg["declines"].to_numpy() / agg["total_auths"].to_numpy(),
51    )
52
53    agg["pct_declines_suspected_fraud"] = np.where(
54        agg["declines"].to_numpy() == 0,
55        0.0,
56        agg["suspected_fraud_declines"].to_numpy() / agg["declines"].to_numpy(),
57    )
58
59    return agg.drop(columns=["suspected_fraud_declines"]).sort_values("date").reset_index(drop=True)
60

1from __future__ import annotations
2
3import pandas as pd
4import numpy as np
5
6
7def add_cbk_30d_rate(transactions: pd.DataFrame) -> pd.DataFrame:
8    """Add cbk_30d_rate per transaction.
9
10    Feature definition for each transaction row:
11      cbk_30d_rate = (# chargebacks in prior 30 days for the account) / (# txns in prior 30 days for the account)
12    Excludes the current transaction from numerator and denominator.
13
14    Expected columns:
15      - account_id
16      - txn_ts
17      - is_chargeback (0/1)
18
19    Returns a copy with a new float column cbk_30d_rate.
20    """
21
22    df = transactions.copy()
23    df["txn_ts"] = pd.to_datetime(df["txn_ts"], utc=True, errors="coerce")
24    df = df.dropna(subset=["txn_ts"])
25
26    # Sort to ensure correct rolling time windows
27    df = df.sort_values(["account_id", "txn_ts"], kind="mergesort").reset_index(drop=True)
28
29    # Work per account using rolling with a time window.
30    def _per_account(g: pd.DataFrame) -> pd.DataFrame:
31        g = g.copy()
32        g = g.sort_values("txn_ts", kind="mergesort")
33        g = g.set_index("txn_ts")
34
35        # Rolling counts over the last 30 days INCLUDING the current timestamp, then shift to exclude current.
36        rolling_txn_cnt = g["is_chargeback"].rolling("30D").count().shift(1)
37        rolling_cbk_cnt = g["is_chargeback"].rolling("30D").sum().shift(1)
38
39        denom = rolling_txn_cnt.to_numpy(dtype=float)
40        numer = rolling_cbk_cnt.to_numpy(dtype=float)
41
42        g["cbk_30d_rate"] = np.where(denom == 0, 0.0, numer / denom)
43
44        g = g.reset_index()
45        return g
46
47    out = df.groupby("account_id", group_keys=False).apply(_per_account)
48
49    # Preserve original column order plus feature at the end
50    if "cbk_30d_rate" not in out.columns:
51        out["cbk_30d_rate"] = 0.0
52
53    return out.reset_index(drop=True)
54

Morgan Stanley's two dedicated SQL rounds feed directly into a case study where you're asked to structure a live fraud investigation, meaning weak querying skills don't just cost you one round, they collapse your case study performance too. The compounding risk most candidates miss is that the statistics questions assume you already know what a chargeback loss rate or wire transfer control actually is, so skipping finance domain prep leaves you unable to even interpret the fraud scenarios you're asked to reason about. If your study plan leans heavily on Python or algorithm puzzles, this distribution is telling you to redirect that time.

Drill Morgan Stanley fraud analytics questions across SQL, statistics, and domain knowledge at datainterview.com/questions.

How to Prepare for Morgan Stanley Data Analyst Interviews

Morgan Stanley is betting big on crypto and digital assets, with a crypto wallet planned for the second half of 2026 and active hiring of lead engineers to build enterprise-grade tokenization infrastructure. That push sits alongside a firm generating $70.3B in annual revenue with 11% year-over-year growth, which means the sheer volume of data flowing through risk, compliance, and client analytics keeps expanding.

When interviewers ask "why Morgan Stanley," don't give a generic prestige answer. Reference something only this firm is doing. You could point to the tokenization buildout and what it means for data infrastructure, or mention their open-source release of CALM (architecture-as-code), which signals a more engineering-forward culture than most banks. Even better, pull a specific figure from the Q4 2025 earnings report and connect it to the kind of analytical work you'd do on the team you're interviewing for.

Try a Real Interview Question

1WITH daily AS (
2  SELECT
3    CAST(t.txn_ts AS DATE) AS txn_date,
4    t.merchant_id,
5    COUNT(*) AS total_txns,
6    SUM(CASE WHEN t.is_fraud = 1 THEN 1 ELSE 0 END) AS fraud_txns
7  FROM transactions t
8  GROUP BY CAST(t.txn_ts AS DATE), t.merchant_id
9), scored AS (
10  SELECT
11    d.txn_date,
12    d.merchant_id,
13    d.total_txns,
14    d.fraud_txns,
15    CAST(d.fraud_txns AS DECIMAL(18,6)) / NULLIF(d.total_txns, 0) AS fraud_rate
16  FROM daily d
17)
18SELECT
19  s.txn_date,
20  s.merchant_id,
21  s.total_txns,
22  s.fraud_txns,
23  s.fraud_rate,
24  DENSE_RANK() OVER (
25    PARTITION BY s.txn_date
26    ORDER BY s.fraud_rate DESC, s.total_txns DESC, s.merchant_id ASC
27  ) AS daily_fraud_rank
28FROM scored s
29ORDER BY s.txn_date, daily_fraud_rank, s.merchant_id;

This type of problem reflects Morgan Stanley's heavy emphasis on SQL across its interview rounds: expect multi-table joins on financial schemas, window functions for running calculations, and questions framed around accounts, transactions, and positions rather than generic web-app tables. Get reps on financial-domain SQL problems at datainterview.com/coding until those patterns feel automatic.

Test Your Readiness

Practice Morgan Stanley-specific interview questions at datainterview.com/questions and weight your study time toward whichever topic areas the widget above shows as most frequent.