Noesis
Level 1 — RecognitionBacktesting Frameworks

Backtesting Frameworks

⏱ 20 minutes40 marksprintable — key stays hidden on paper

Subject: Stock-Market | Chapter: 6.2 Backtesting Frameworks Level: 1 — Recognition Time Limit: 20 minutes Total Marks: 40

Section A — Multiple Choice (1 mark each, 10 marks)

Choose the single best answer.

Q1. Survivorship bias in a backtest occurs when the dataset:

  • A) Includes too many small-cap stocks
  • B) Excludes companies that were delisted or went bankrupt
  • C) Contains duplicate price entries
  • D) Uses daily instead of intraday data

Q2. Look-ahead bias is best described as:

  • A) Using information in a decision that would not have been available at that point in time
  • B) Testing a strategy only on future data
  • C) Assuming zero transaction costs
  • D) Trading too frequently

Q3. CAGR stands for:

  • A) Compound Average Growth Rate
  • B) Compound Annual Growth Rate
  • C) Cumulative Annual Gross Return
  • D) Continuous Average Gross Return

Q4. Which Python library is a dedicated event-driven backtesting framework?

  • A) NumPy
  • B) Matplotlib
  • C) Backtrader
  • D) Requests

Q5. Slippage refers to:

  • A) The commission charged by a broker
  • B) The difference between expected execution price and actual fill price
  • C) The tax on capital gains
  • D) The bid-ask spread only

Q6. Out-of-sample testing is used primarily to:

  • A) Speed up computation
  • B) Detect overfitting by validating on unseen data
  • C) Reduce transaction costs
  • D) Increase the number of trades

Q7. Maximum drawdown measures:

  • A) The largest single-day loss
  • B) The largest peak-to-trough decline in portfolio value
  • C) The total number of losing trades
  • D) The average return per trade

Q8. Paper trading is best defined as:

  • A) Trading only in low-liquidity stocks
  • B) Simulated live trading with no real capital at risk
  • C) Backtesting on historical data
  • D) Trading using printed order forms

Q9. Monte Carlo simulation in backtesting is mainly used to:

  • A) Guarantee future returns
  • B) Assess the range/distribution of possible outcomes via randomization
  • C) Remove survivorship bias
  • D) Source clean historical data

Q10. Clean historical data should ideally be:

  • A) Adjusted for splits and dividends, and free of gaps/errors
  • B) Only closing prices with no volume
  • C) Rounded to whole numbers
  • D) Limited to one year

Section B — Matching (1 mark each, 8 marks)

Q11–Q18. Match each term in Column X to its correct description in Column Y. Write the letter.

# Column X
Q11 Survivorship bias
Q12 Look-ahead bias
Q13 Slippage
Q14 Transaction cost
Q15 In-sample data
Q16 Max drawdown
Q17 CAGR
Q18 Paper trading
Letter Column Y
A Commissions/fees deducted per trade
B Data used to build/optimize the strategy
C Simulated real-time trading with no real money
D Excluding failed firms from the dataset
E Peak-to-trough portfolio decline
F Using future data unavailable at decision time
G Annualized smoothed rate of return
H Gap between expected and actual fill price

Section C — True/False with Justification (2 marks each, 22 marks)

State True or False (1 mark) and give a one-line justification (1 mark).

Q19. Ignoring transaction costs makes a backtest look more profitable than it would be in reality.

Q20. Survivorship bias tends to understate a strategy's historical returns.

Q21. Using the closing price of a day to trigger a trade executed at that same day's open is a form of look-ahead bias.

Q22. A strategy that performs well in-sample is guaranteed to perform well out-of-sample.

Q23. Higher slippage assumptions make a backtest more conservative.

Q24. Max drawdown of 40% means the portfolio lost 40% of its value from a peak before recovering.

Q25. Monte Carlo simulation can prove exactly what a strategy will return next year.

Q26. pandas can be used to compute returns and drawdowns but is not itself an event-driven backtest engine.

Q27. Paper trading eliminates all differences between simulated and live performance.

Q28. A CAGR of 10% over 3 years means the investment grew by a total of exactly 30%.

Q29. Dividend and split adjustments are part of preparing clean historical data.

Answer keyMark scheme & solutions

Section A (1 mark each)

Q Ans Why
Q1 B Survivorship bias = dropping delisted/bankrupt firms, inflating returns.
Q2 A Look-ahead = using info not yet available at decision time.
Q3 B Compound Annual Growth Rate.
Q4 C Backtrader is a dedicated event-driven backtesting library.
Q5 B Slippage = expected vs actual fill price difference.
Q6 B OOS testing validates on unseen data to catch overfitting.
Q7 B Max DD = largest peak-to-trough decline.
Q8 B Paper trading = simulated live trading, no real capital.
Q9 B Monte Carlo produces a distribution of possible outcomes.
Q10 A Clean data is split/dividend adjusted and gap/error free.

Section B (1 mark each)

Q Ans Q Ans
Q11 D Q15 B
Q12 F Q16 E
Q13 H Q17 G
Q14 A Q18 C

Section C (1 mark T/F + 1 mark justification)

Q19. True — Real costs reduce net returns; ignoring them overstates profit.

Q20. False — Survivorship bias overstates returns because only successful firms remain.

Q21. True — At the open you cannot yet know that day's close; using it is look-ahead bias.

Q22. False — In-sample success may be overfitting; OOS is required to confirm.

Q23. True — Larger slippage lowers modeled fills/returns, making results more conservative/realistic.

Q24. True — 40% drawdown = 40% drop from a prior peak. (Note: recovery is not required for a drawdown to count, but the statement remains true.)

Q25. False — Monte Carlo gives probabilistic ranges, not exact predictions.

Q26. True — pandas is a data/analytics library; engines like backtrader/zipline handle event-driven simulation.

Q27. False — Paper trading omits real slippage, liquidity, psychology, latency differences.

Q28. False — 10% CAGR over 3 years compounds to 1.13−1≈33.1%1.1^3-1 \approx 33.1\%, not 30%.

Q29. True — Adjusting for corporate actions is a core data-cleaning step.

Worked calculation for Q28

Total growth=(1+0.10)3−1=1.331−1=0.331=33.1%\text{Total growth} = (1+0.10)^3 - 1 = 1.331 - 1 = 0.331 = 33.1\% So it is not exactly 30% — statement is False.

CAGR reference (for context)

CAGR=(VendVstart)1/n−1\text{CAGR} = \left(\frac{V_{end}}{V_{start}}\right)^{1/n} - 1

[
  {"claim":"Q28: 10% CAGR over 3 years gives ~33.1% total, not 30%","code":"total = (1 + Rational(1,10))**3 - 1; result = (total == Rational(331,1000)) and (total != Rational(3,10))"},
  {"claim":"Q24: a 40% drawdown corresponds to multiplier 0.6","code":"peak = 100; trough = peak*(1 - Rational(40,100)); result = (trough == 60)"},
  {"claim":"CAGR formula: doubling in 2 years => ~41.4% CAGR","code":"cagr = (Integer(2))**(Rational(1,2)) - 1; result = simplify(cagr - (sqrt(2)-1)) == 0"}
]