Thanks for this incredibly thoughtful comment and the BOTEC work! It is exciting to see someone digging into the implications for the WELLBY impact.
Here are my thoughts on your questions:
1. ISA Recovery & Cost per Person You are right that the cost-effectiveness depends heavily on the recycling of funds. I built two related Monte Carlo simulations to assess the sensitivity of IRR, impact, and students educated to our input parameters. For the sake of brevity, I won’t paste the full methodology here (it helps explain the logic behind the sheer number of variables), but I’ve attached a takeaway screenshot below. It shows how many students would be supported over the next 55 years with $1m in recycling investment.
The assumptions in our GiveWell sheet align with the p25 Scenario, which could be interpreted (539 student on an initial $1m investment, recycled) as a ~$2k cost per student.
I want to caveat that this has optimistic and pessimistic assumptions. There are some indivisibilities that could make expenses higher if the program runs at a smaller scale than a budget (explained below) of around $5m a year.
On the other hand, this represents completely unleveraged investment. With a stable underwriting model, Malengo (or similar lending with an educational migration approach) could eventually be lending against the contracts. A reasonable rule of thumb, if we are successful, would be achieving 5:1 leverage. In that scenario, $1m in the first-loss guarantee would unlock $5m in capital that would send at least 5x as many students, knocking the cost per student to $400. (Note: We might be able to swing something like 2:1 leverage at the moment, but it is a lot of work!)
In the true optimistic case, Malengo has infrastructure-like financing where donors wouldn’t even need to donate to Malengo, they only offer a guarantee backed by assets in their DAF or Foundation or it is offered by a DFI and that unlocks bank loans that are serviced by the ISA obligations. In this scenario, philanthropy covers the fixed cost of building the “flywheel”, while the scaling capital itself comes from credit markets.
I am hoping that our new CIO, Chad Sterbenz, will be able to post to the EA forum soon to do more justice to how this type of fund would work in practice with examples.
2. Cost Duration To answer your clarification: Yes, the model accounts for 1-year stipend and then wrap-around support through the full duration of schooling. The “cost per person” is an all-in figure. It is derived from a conservative strategic plan: it represents the total dollars spent per student to sustain the entire organization for 20 years at a minimum viable scale (covering tuition, living cost gaps, and operational overhead), with a buffer in case a wind-down is required.
3. Disaggregated SWB/MH Measures This is a great point about the difference between life satisfaction and affective happiness. I don’t have that breakdown on hand, but I believe you are correct about the divergence. I will inquire with the research team about sharing those exact measures.
Regarding your BOTEC assumptions: Your spillover duration assumption (pegged to parents’ lives) strikes me as reasonable, though perhaps conservative in scope. In practice, scholars send remittances to many family members, particularly to fund younger siblings’ schooling. (We don’t currently expect them to facilitate additional migration via their own funds). I actually incorporated a taper reflecting this into the impact simulation shared above after reading your comment. Thank you for that nudge!
Second, regarding the assumption that wellbeing benefits represent a flat line: I agree that is conservative. If the economic integration works as intended (see point 4), we should expect the convergence toward host-country wellbeing levels to be faster and stickier for these scholars than for the average migrant.
4. The Political Question & The OECD Graph You anticipated the rejoinder perfectly. I’m with Alexander Kustov on this: migration has to be demonstrably beneficial to receiving countries. Migration will happen either through chaos or competence. Money can buy competence if the contract lets it.
The ISA is that contract: it pays for language, skills, and placement quality, and then pays itself back. We are betting that competence and successful integration are the best counter-arguments to populism.
As shown in your OECD chart, there is a massive gap in Germany between immigrants with foreign degrees and those with host-country degrees.
By ensuring our scholars get German degrees, we bypass the primary friction point. Our scholars face a lower “immigrant penalty” in the labor market sense; they (hopefully) perform almost identically to native-born graduates (We model 20% discount decaying as they integrate for 10 years of working). Malengo wants to demonstrate that deep economic integration can happen with sufficient resources.
I’m going to focus on the funds recycling since this seems like it may be the most important bit.
Since you’re assuming $1mil recycled for a budget of $5mil does this imply that you will recycle 20% of costs? If so, that seems low given everything else you’ve said, so I assume I’m missing something.
The leverage point is really interesting. Do you have any current prospects for making the leverage work? I would guess you have to wait (how long?) to see if the default rate is sufficiently low for the ISA?
What has to be true for this to work out financially? Something like the real rate of return on $ / euro invested in students has to be equal or greater than the interest rate on the loan?
It’d be great to hear more about this!
When you mention $2k and $400 per student, could you explain how this differs from the $30k listed as the best guess in the spreadsheet. Sorry for all the questions, I just want to understand!
The mechanics of the fund are counter-intuitive compared to standard grant-making, so let me break down the math behind the $2k and $400 figures and how they relate to the $30k cost.
1. The Recycling vs. The Multiplier Effect I think the confusion here stems from a coincidence of numbers in my previous comment. I mentioned a $5m budget and a $5m leverage target, but those are distinct concepts.
The Recycling Math (p25 Scenario): In this scenario, we start with $1m in philanthropic capital and invest at a gross cost of 30k per student, that sends ~32 students immediately.
Over the next 40 years, as those students repay, the money is lent out again. Each student funds about 1.5 new students every 10 years (ie 150% recovery). In the simulation, that single $1m pot eventually funds a total of 539 students.
This implies a multiplier of roughly 16x: For every 1 student funded by the initial donation, the returns eventually fund ~15 more.
$30k (Gross Cost): This is the “sticker price.” It is the total cash budgeted to send one student (tuition + living stipend + operational overhead) → given our small size stays at this size.
~$2k (Net Philanthropic Cost): This is the unrecovered cost per student in a 40-year period. If $1m of philanthropy eventually educates 539 students (via the recycling described above), the effective cost to the donor per student is $1m / 539 = $1,855.
~$400 (Leveraged Cost): This assumes we achieve the 5:1 leverage ratio. If a donor provides the $1m philanthropic capital and we build a fund with it as first-loss capital, and we raise $5m in commercial debt, the system has $5m to deploy ($1m in reserve). The donor’s “cost per student” drops because their single dollar unlocked five dollars of capacity. → Note, this is quite simplified because we become much more sensitive to assumptions while leveraged. Returns are either much greater or the risk capital is all lost.
2. Financial Viability & Blended Capital You asked: “Something like the real rate of return on $ / euro invested in students has to be equal or greater than the interest rate on the loan?”
Yes, exactly. In nominal terms, the Net Portfolio Yield (after defaults/expenses) must be greater than Weighted Average Cost of Capital (WACC).
In the p25 scenario, the Net Portfolio Yield is 3.8% (Listed under the IRR column).
This is too low to attract pure commercial capital (Even with data, European rates for this risk profile might be 4–7%). However, this is where Blended Finance comes in. We don’t need 100% commercial capital; we can stack different tranches:
Tranche A (Philanthropy/DFI): 33% of the fund. Target return: 0%.
Tranche B (Commercial): 66% of the fund. Target return: ~6%.
Result: The blended Cost of Capital drops to ~4%, which makes the 3.8% return viable (or very close to it).
I have initiated talks with impact funds, DFIs, and banks to structure this. It is difficult to achieve, but within reach, so we hired a CIO. While lenders naturally want years of repayment data, there are creative ways to use philanthropic guarantees to de-risk them early on. We hope to share a more detailed write-up on the structuring of these capital stacks soon!
This is super helpful to have explained and makes more sense now.
So what is your best guess of the cost per student educated given Malengo’s expected ability to recycle costs and leverage? What is the cost per person Malengo educates / facilitates the immigration of that you would put into my BOTEC?
I am comfortable recommending a rule that is internally valid with all our presented assumptions. Our analysis operates in the 25th percentile of outcomes; that implies a cost per student of $1,855. (This lines up with the ~95% cost coverage in your row 39, though that is a happy coincidence!)
However, if you choose to weight across the distribution, here is how the cost per student evolves based on recycling performance and leverage assumptions:
p10: $3,472 (Zero leverage)
p25: $1,855 (Zero leverage)
p50: $353 (Assuming 2:1 leverage)
p75: $377 (Assunming 2:1 leverage; note: more Master’s students in this scenario compensates for lower dropout rates, leading to the flatness here)
p90: $141 (5:1 leverage)
Note: These figures are derived from the distribution in the top comment: $1m / (num_students * leverage).
If you want my personal “best guess”: I believe management will react to the data. If we are able to iterate for a decade, we will push toward p90. The team will find cost reductions using partners and tech, optimize contract specifications to ensure we achieve the target leverage, and refine the underwriting model to find the students most likely to succeed.
So, for your BOTEC, I would put $141 per student (implying ~99.5% effective cost coverage via recycling + leverage). But bear in mind, you are speaking to the person who already made that wager!
Thanks for this incredibly thoughtful comment and the BOTEC work! It is exciting to see someone digging into the implications for the WELLBY impact.
Here are my thoughts on your questions:
1. ISA Recovery & Cost per Person
You are right that the cost-effectiveness depends heavily on the recycling of funds. I built two related Monte Carlo simulations to assess the sensitivity of IRR, impact, and students educated to our input parameters. For the sake of brevity, I won’t paste the full methodology here (it helps explain the logic behind the sheer number of variables), but I’ve attached a takeaway screenshot below. It shows how many students would be supported over the next 55 years with $1m in recycling investment.
The assumptions in our GiveWell sheet align with the p25 Scenario, which could be interpreted (539 student on an initial $1m investment, recycled) as a ~$2k cost per student.
https://malengo.org/impact_simulation/
I want to caveat that this has optimistic and pessimistic assumptions. There are some indivisibilities that could make expenses higher if the program runs at a smaller scale than a budget (explained below) of around $5m a year.
On the other hand, this represents completely unleveraged investment. With a stable underwriting model, Malengo (or similar lending with an educational migration approach) could eventually be lending against the contracts. A reasonable rule of thumb, if we are successful, would be achieving 5:1 leverage. In that scenario, $1m in the first-loss guarantee would unlock $5m in capital that would send at least 5x as many students, knocking the cost per student to $400. (Note: We might be able to swing something like 2:1 leverage at the moment, but it is a lot of work!)
In the true optimistic case, Malengo has infrastructure-like financing where donors wouldn’t even need to donate to Malengo, they only offer a guarantee backed by assets in their DAF or Foundation or it is offered by a DFI and that unlocks bank loans that are serviced by the ISA obligations. In this scenario, philanthropy covers the fixed cost of building the “flywheel”, while the scaling capital itself comes from credit markets.
I am hoping that our new CIO, Chad Sterbenz, will be able to post to the EA forum soon to do more justice to how this type of fund would work in practice with examples.
2. Cost Duration
To answer your clarification: Yes, the model accounts for 1-year stipend and then wrap-around support through the full duration of schooling. The “cost per person” is an all-in figure. It is derived from a conservative strategic plan: it represents the total dollars spent per student to sustain the entire organization for 20 years at a minimum viable scale (covering tuition, living cost gaps, and operational overhead), with a buffer in case a wind-down is required.
3. Disaggregated SWB/MH Measures
This is a great point about the difference between life satisfaction and affective happiness. I don’t have that breakdown on hand, but I believe you are correct about the divergence. I will inquire with the research team about sharing those exact measures.
Regarding your BOTEC assumptions: Your spillover duration assumption (pegged to parents’ lives) strikes me as reasonable, though perhaps conservative in scope. In practice, scholars send remittances to many family members, particularly to fund younger siblings’ schooling. (We don’t currently expect them to facilitate additional migration via their own funds). I actually incorporated a taper reflecting this into the impact simulation shared above after reading your comment. Thank you for that nudge!
Second, regarding the assumption that wellbeing benefits represent a flat line: I agree that is conservative. If the economic integration works as intended (see point 4), we should expect the convergence toward host-country wellbeing levels to be faster and stickier for these scholars than for the average migrant.
4. The Political Question & The OECD Graph
You anticipated the rejoinder perfectly. I’m with Alexander Kustov on this: migration has to be demonstrably beneficial to receiving countries. Migration will happen either through chaos or competence. Money can buy competence if the contract lets it.
The ISA is that contract: it pays for language, skills, and placement quality, and then pays itself back. We are betting that competence and successful integration are the best counter-arguments to populism.
As shown in your OECD chart, there is a massive gap in Germany between immigrants with foreign degrees and those with host-country degrees.
By ensuring our scholars get German degrees, we bypass the primary friction point. Our scholars face a lower “immigrant penalty” in the labor market sense; they (hopefully) perform almost identically to native-born graduates (We model 20% discount decaying as they integrate for 10 years of working). Malengo wants to demonstrate that deep economic integration can happen with sufficient resources.
Thank you for the response Richard!
I’m going to focus on the funds recycling since this seems like it may be the most important bit.
Since you’re assuming $1mil recycled for a budget of $5mil does this imply that you will recycle 20% of costs? If so, that seems low given everything else you’ve said, so I assume I’m missing something.
The leverage point is really interesting. Do you have any current prospects for making the leverage work? I would guess you have to wait (how long?) to see if the default rate is sufficiently low for the ISA?
What has to be true for this to work out financially? Something like the real rate of return on $ / euro invested in students has to be equal or greater than the interest rate on the loan?
It’d be great to hear more about this!
When you mention $2k and $400 per student, could you explain how this differs from the $30k listed as the best guess in the spreadsheet. Sorry for all the questions, I just want to understand!
The mechanics of the fund are counter-intuitive compared to standard grant-making, so let me break down the math behind the $2k and $400 figures and how they relate to the $30k cost.
1. The Recycling vs. The Multiplier Effect
I think the confusion here stems from a coincidence of numbers in my previous comment. I mentioned a $5m budget and a $5m leverage target, but those are distinct concepts.
The Recycling Math (p25 Scenario): In this scenario, we start with $1m in philanthropic capital and invest at a gross cost of 30k per student, that sends ~32 students immediately.
Over the next 40 years, as those students repay, the money is lent out again. Each student funds about 1.5 new students every 10 years (ie 150% recovery). In the simulation, that single $1m pot eventually funds a total of 539 students.
This implies a multiplier of roughly 16x: For every 1 student funded by the initial donation, the returns eventually fund ~15 more.
$30k (Gross Cost): This is the “sticker price.” It is the total cash budgeted to send one student (tuition + living stipend + operational overhead) → given our small size stays at this size.
~$2k (Net Philanthropic Cost): This is the unrecovered cost per student in a 40-year period. If $1m of philanthropy eventually educates 539 students (via the recycling described above), the effective cost to the donor per student is $1m / 539 = $1,855.
~$400 (Leveraged Cost): This assumes we achieve the 5:1 leverage ratio. If a donor provides the $1m philanthropic capital and we build a fund with it as first-loss capital, and we raise $5m in commercial debt, the system has $5m to deploy ($1m in reserve). The donor’s “cost per student” drops because their single dollar unlocked five dollars of capacity. → Note, this is quite simplified because we become much more sensitive to assumptions while leveraged. Returns are either much greater or the risk capital is all lost.
2. Financial Viability & Blended Capital
You asked: “Something like the real rate of return on $ / euro invested in students has to be equal or greater than the interest rate on the loan?”
Yes, exactly. In nominal terms, the Net Portfolio Yield (after defaults/expenses) must be greater than Weighted Average Cost of Capital (WACC).
In the p25 scenario, the Net Portfolio Yield is 3.8% (Listed under the IRR column).
This is too low to attract pure commercial capital (Even with data, European rates for this risk profile might be 4–7%). However, this is where Blended Finance comes in. We don’t need 100% commercial capital; we can stack different tranches:
Tranche A (Philanthropy/DFI): 33% of the fund. Target return: 0%.
Tranche B (Commercial): 66% of the fund. Target return: ~6%.
Result: The blended Cost of Capital drops to ~4%, which makes the 3.8% return viable (or very close to it).
I have initiated talks with impact funds, DFIs, and banks to structure this. It is difficult to achieve, but within reach, so we hired a CIO. While lenders naturally want years of repayment data, there are creative ways to use philanthropic guarantees to de-risk them early on. We hope to share a more detailed write-up on the structuring of these capital stacks soon!
This is super helpful to have explained and makes more sense now.
So what is your best guess of the cost per student educated given Malengo’s expected ability to recycle costs and leverage? What is the cost per person Malengo educates / facilitates the immigration of that you would put into my BOTEC?
At the risk of being pedantic, the answer depends heavily on your decision rule regarding uncertainty (relevant: Noah Haber’s work on GiveWell’s uncertainty problem).
I am comfortable recommending a rule that is internally valid with all our presented assumptions. Our analysis operates in the 25th percentile of outcomes; that implies a cost per student of $1,855. (This lines up with the ~95% cost coverage in your row 39, though that is a happy coincidence!)
However, if you choose to weight across the distribution, here is how the cost per student evolves based on recycling performance and leverage assumptions:
p10: $3,472 (Zero leverage)
p25: $1,855 (Zero leverage)
p50: $353 (Assuming 2:1 leverage)
p75: $377 (Assunming 2:1 leverage; note: more Master’s students in this scenario compensates for lower dropout rates, leading to the flatness here)
p90: $141 (5:1 leverage)
Note: These figures are derived from the distribution in the top comment: $1m / (num_students * leverage).
If you want my personal “best guess”: I believe management will react to the data. If we are able to iterate for a decade, we will push toward p90. The team will find cost reductions using partners and tech, optimize contract specifications to ensure we achieve the target leverage, and refine the underwriting model to find the students most likely to succeed.
So, for your BOTEC, I would put $141 per student (implying ~99.5% effective cost coverage via recycling + leverage). But bear in mind, you are speaking to the person who already made that wager!
I appreciate the caveat and you sharing your best guess!