3-4 Flashcards
(10 cards)
What is the analogous cost estimation’s cost estimating relationship (CER)?
Analogous cost estimation uses the cost of a similar past project to estimate a new project’s cost.
CER is the simple formula:
Cost per unit from the old project × Size of new project = Estimated cost of new project.
Example:
Old project: $100,000 for 10 turbines → $10,000 per turbine
New project: 15 turbines → $10,000 × 15 = $150,000
Simple Explanation:
CER means using the cost per unit from a past project to guess the cost for a new, similar project by multiplying by how big the new project is.
What are the steps involved in doing an analogous cost estimation?
Pick a past, similar project.
- Choose one key cost driver (e.g., number of units).
- Plot the past project cost against its cost driver (e.g., 10 units, $100,000).
- Assume cost is zero when the cost driver is zero (0,0).
- Draw a straight line from (0,0) to the past project point.
- Use the slope (cost per unit) × new project size to estimate cost.
Simple Explanation:
Find one similar project, use its cost and size to find a unit cost, then multiply that by the new project’s size to estimate its cost.
What are the assumptions made when using analogous cost estimation?
Linear scaling: Cost scales linearly with the selected cost driver.
If 10 units cost $100k, then 20 should cost $200k.
Zero intercept: A project with a cost driver value of 0 would cost 0.
Single cost driver: Only one variable explains the variation in cost.
You’re ignoring other things that could affect the cost.
Similarity: The chosen historical project is sufficiently similar to the new project.
This is critical—if the projects are too different, the estimate is unreliable.
No adjustment for noise: The estimation inherently includes both signal and noise from the single observation.
Whatever randomness or unique things happened in the old project will affect the estimate too, because you’re using only one data point.
Simple Explanation:
It assumes cost grows steadily with size, only one factor matters, and the old project is very similar to the new one.
What is the relationship between analogous cost estimation and simple cross-sectional ordinary least squares regression?
Analogous cost estimation is a special case of ordinary least squares (OLS) regression:
It's like doing OLS regression with only one data point, and forcing the line through the origin (0,0).
In OLS, you use many data points to find the "best fit" line.
In analogous estimation, you use just one point, so your "line" is fully defined by:
The origin (0,0)
The one data point from a past project
What are the Weaknesses of Analogous Cost Estimation
- Only one cost driver
Real projects are affected by many factors. Ignoring the rest can make the estimate very wrong. - Forces zero intercept
Most real-world projects cost something even if you build very little. Forcing the line through (0,0) makes the slope too steep and can under/overestimate. - Only one observation
You’re copying everything about that one project—including random quirks and mistakes. - No error correction
OLS regression tries to average out noise. Analogous estimation doesn’t—it just scales whatever noise was in the original data point.
Analogous estimation is like:
“Hey, the last time we built a bridge that was 100 meters long, it cost $1 million. Now we’re building one that’s 150 meters. So it’ll probably cost $1.5 million.”
It’s quick, dirty, and sometimes good enough. But it relies on a lot of faith in the similarity and quality of the old project.
If you had more data, you’d use proper regression analysis to get better results.
What is Monte Carlo Simulation (MCS) in Project Cost Estimation?
Monte Carlo Simulation estimates the uncertainty of project costs by generating many possible outcomes instead of one fixed number.
Inputs are ranges or probability distributions (e.g., material cost $30–50M, labor $40–60M). Random samples from these distributions are combined thousands of times. The results form a probability distribution (e.g., histogram) showing likely costs and risks.
Example outcome:
Most likely cost: $90M
10% chance cost > $100M
Range: $72M to $108M
Simple Explanation:
MCS helps see how costs might change by simulating many possible scenarios, giving a full picture of risk.
How Do We Generate Random Numbers for MonteCarloS?
Computers generate uniform random numbers between 0 and 1 easily.
To get random values following other distributions (like normal or beta), use:
Inverse Cumulative Distribution Function (Inverse CDF or quantile function)
Input: a uniform random number (e.g., 0.84) Output: a value that follows the desired distribution (e.g., 0.99 on a normal curve)
Simple Explanation:
Start with simple random numbers and convert them into realistic values using math functions to match real-world patterns.
Flashcard 3: What is a Pseudo-Random Number Generator (PRNG)? What Makes a Good One?
A PRNG is an algorithm that produces sequences of numbers that appear random but are deterministic, based on an initial seed.
Good PRNG qualities:
Uniform distribution (evenly spread numbers) Independence (no patterns between numbers) Long period (doesn't repeat for a very long time) Reproducible if seed is known, unpredictable otherwise
Example: Linear Congruential Generator (LCG).
Simple Explanation:
PRNGs make “random-like” numbers for simulations, which need to look random, not repeat quickly, and be reliable.
How would u design a Monte Carlo Simulation?
Define Goal: Estimate Net Present Value (NPV) profitability of solar panels.
Identify Variables: Electricity price, production capacity, costs, subsidies.
Assign Distributions: Electricity price ~ Normal, cost overruns ~ Beta, etc.
Build Model: Calculate yearly cash flows and total NPV using input variables.
Run Simulation: Repeat calculation many times, sampling random values from distributions.
Analyze Results: Average NPV, probability of loss, risk percentiles, visualize results.
Simple Explanation:
Model uncertain inputs with probability, simulate many scenarios, then analyze profit chances and risks.
What is a PERT Variable?
How to Simulate Two Correlated PERT Variables
A PERT variable is used to model uncertain values when you have:
a best-case (optimistic), a most likely, and a worst-case (pessimistic) estimate.
PERT variables use three estimates (best, most likely, worst) to model uncertain outcomes smoothly. To simulate two that are related (correlated), you:
Start with two random numbers. Adjust them so they move partly together. Plug them into the PERT model to get realistic, linked outcomes.
This lets you model things like “when one goes up, the other probably does too.”
