What Your Lease Portfolio Looks Like Through a Markowitz Lens

Applying modern portfolio theory to corporate real estate--where the goal isn't higher returns, but lower cost at less risk

Anyone who manages a portfolio of assets knows not to put everything into one investment. In the leasing world, on the other hand, we tend to view each deal on its own terms and with an almost complete focus on cost and time. To the extent that portfolio analysis is performed, it frequently centers on lease expirations viewed over some finite term together with the critical issue of user demand.

Large corporate occupiers usually have large real estate footprints which represent a significant percentage of corporate expense. When considering these portfolios, could one apply some of the same tools used by managers of assets--or analogs of those tools--to craft strategies that quantify and optimize lease portfolios in terms of cost and risk?

The short answer is yes. And the tools have been available since 1952, when Harry Markowitz published “Portfolio Selection” and launched what we now call Modern Portfolio Theory. In simple terms, Markowitz showed that a portfolio made of assets that co-vary imperfectly can improve the risk-return tradeoff: such a structured portfolio can maintain the same return with lower risk, or maintain the same risk with higher returns. Every wealth manager, pension fund, and endowment on the planet uses some version of this framework. Commercial real estate, despite being one of the largest line items on a corporate balance sheet, has largely ignored it.

This article shows some ways to think about occupier portfolios at scale that builds on work I did in a 2008 Real Estate Review article, where I used Monte Carlo simulation to study a landlord’s lease portfolio--diversified across tenants of different sizes and credit qualities. That article showed that combining assets with imperfect correlation could reduce portfolio risk in the context of real estate, the same result Markowitz demonstrated for securities. In a companion article, I applied bond market duration analysis to an occupier’s lease portfolio, showing how the tools of fixed-income portfolio management could inform lease structuring decisions.

The 2008 article studied the risk of tenant default and the time it took to lease a building, both landlord’s concerns. This article looks at an occupier’s cost of space--across dozens of leases in multiple markets--and looks at both cost and volatility of outcomes. The factor at risk is the exposure created by leases with different terms to different amounts of market uncertainty. Shorter-term leases are exposed to spikes in market rents; longer-term leases are exposed to the steady growth of market rents over time. And the geographic mix determines how these exposures interact.

Every lease event--renewal, sublease, relocation--is a portfolio trade. Most occupiers treat these as isolated decisions. This framework says: they’re not.

The Math: Geometric Brownian Motion and Monte Carlo Simulation

Two mathematical concepts underpin the analysis. Neither is new, and neither requires much math to understand intuitively.

The first is Geometric Brownian Motion, or GBM. This is the standard model for how asset prices evolve over time under uncertainty. In our context, it describes how market rents in each city might grow (or shrink) over say the next decade. The GBM model has two parameters: a drift rate (the expected annual rent growth) and a volatility (the standard deviation of that growth). Together, these produce rent paths that trend upward on average but wander unpredictably around that trend.

Think of it this way: if New York office rents have an expected growth rate of 3% per year and a volatility of 12%, then next year’s rent is probably around 3% higher--but it could be 15% higher or 9% lower. Over ten years, the paths fan out like branches of a tree. The trunk is the expected trajectory. The branches are the uncertainty.

The second concept is Monte Carlo simulation. Instead of trying to solve for one future analytically, Monte Carlo is used here to simulate a thousand of them. Each trial draws a random rent path for all 18 markets simultaneously, respecting some level of correlation between markets. Then we compute the total occupancy cost for the portfolio under each scenario. The output isn’t a single number--it’s a distribution with an expected cost (the mean across trials) and a cost volatility (the standard deviation).

The critical detail is that markets are correlated but not identical. New York and San Francisco tend to move together--when coastal rents spike, they spike in both cities. The correlation between those two markets is about 0.80. But the pattern is not as simple as “all big cities move together.” Chicago, despite being a major market, has almost zero correlation with New York--its steady, low-volatility rent growth looks nothing like New York’s boom-bust cycles. Some pairs are actually negatively correlated: when New York rents rise, Phoenix rents tend to fall, and vice versa. This imperfect--and sometimes inverse--correlation is what makes diversification work. If every market moved in lockstep, no amount of reshuffling would reduce risk. However, they don’t and it does.

The Data

To do this analysis a program was developed using Claude Code and Python which created a synthetic portfolio of 250 leases spread across 18 US markets--New York, San Francisco, Boston, Los Angeles, Chicago, Washington DC, Seattle, and eleven secondary and tertiary markets including Dallas, Atlanta, Minneapolis, Charlotte, Nashville, Phoenix, Denver, and others. The distributions are calibrated to public market data from CoStar and CBRE benchmarks, but no proprietary information was used. Think of it as a controlled laboratory: we can change one variable and hold everything else fixed, which is something you can never do with a real portfolio.

The leases span three building classes (A, B, and C), with original terms ranging from three to fifteen years. Some have flat rents, some escalate at 2.5% or 3% annually. It’s a messy, realistic collection--the kind of portfolio a CRE team with 200+ locations might actually manage.

Of the 250 leases, 192 are active with meaningful remaining term. Here are the basics:

That last point--the concentration in high-rent markets--is where the analysis gets interesting. Two-thirds of your portfolio cost sits in a handful of high-rent cities. Some of those cities are highly correlated with each other (New York and San Francisco move almost in lockstep), while others are not (Chicago’s rent growth bears almost no resemblance to New York’s). The question isn’t just how much you’re spending--it’s how much of your cost volatility is coming from concentrated, correlated exposure.

A note on feasibility. Broad-based simulation of large lease portfolios by hand (and by hand I include Excel) has been the practical ceiling for most analysts. Running a thousand Monte Carlo trials across 250 leases in 18 correlated markets, then computing covariance matrices and optimizing frontiers, is not a spreadsheet exercise. However, the combination of AI, relational databases, and Python makes this type of analysis, if not quite straightforward, then at least available to people conversant with these tools in their current 2026 incarnation. The entire simulation engine behind this article--data generation, stochastic modeling, Markowitz optimization, and chart production--runs in a few minutes on a laptop. The barrier is no longer computational. It’s conceptual.

From Yield to Cost

In Markowitz’s original framework, the investor holds a portfolio of assets and wants to maximize expected return for a given level of risk. For a corporate occupier, the objective is reversed: you hold a portfolio of cost obligations and want to minimize expected cost for a given level of cost volatility.

The math is identical. Only the sign changes. Instead of plotting expected return on the Y-axis and standard deviation of returns on the X-axis, we plot expected cost per square foot against standard deviation of cost per square foot. But there’s a visual trick worth noting: because lower cost is better (unlike higher return), we invert the vertical axis--putting lower cost at the top. This preserves the standard MPT convention where “better” is up and to the left. The efficient frontier--the curve that represents the best achievable tradeoff--still has the same hyperbolic shape that Markowitz drew by hand seventy years ago.

What does “cost volatility” mean concretely? Across 1,000 simulated futures, each with different rent paths for all 18 markets, the total occupancy cost of the portfolio varies. In some scenarios rents spike in your major markets and cost is high. In others they stay flat or decline. The standard deviation of cost per square foot across those 1,000 outcomes is the portfolio’s risk. A portfolio concentrated in volatile markets has a wide spread of outcomes--high risk. A portfolio diversified across imperfectly correlated markets has a narrower spread--lower risk.

The covariance between markets comes from historical rent growth data. For each pair of markets, we measure how their annual rent changes have moved together over the past 15 to 18 years. Some pairs co-move strongly (New York and San Francisco), some barely at all (New York and Dallas), and some move in opposite directions (New York and Phoenix). This covariance matrix--18 by 18--is the raw material for the frontier computation.

How Markets Vary Together

The geographic frontier tells you which allocations are efficient, but it doesn’t explain why diversification works here. That requires one additional concept: covariance.

Covariance measures how two things move together. Formally, the covariance between markets i and j is:

where ρ(i,j) is the correlation between market i and market j (a number between -1 and +1), and σ(i) and σ(j) are the respective rent volatilities. Correlation is just covariance divided by the product of the two standard deviations--it normalizes the number so you can compare pairs regardless of scale.

To estimate these correlations, we extracted annual average asking rent data from Newmark Research quarterly reports for 11 U.S. markets, covering 2007 through 2025. The data is approximate--rent levels were back-calculated from labeled year-over-year growth rates anchored to confirmed recent values--but the growth rates themselves are precise, and it is the correlations between growth rates that matter for this analysis.¹

New York and San Francisco are highly correlated at 0.81--coastal tech-and-finance hubs that boom and bust together. But Chicago has a correlation of -0.15 with New York. Chicago’s rent growth is steady and low-volatility (standard deviation of 1.8% per year, compared to New York’s 7.6%), and it moves to a completely different rhythm. Los Angeles correlates at 0.52 with New York--moderate. Washington DC correlates with New York at essentially zero (0.02).

The most striking finding is that several market pairs are negatively correlated. New York and Phoenix have a correlation of -0.45. San Francisco and Phoenix: -0.58. When coastal rents spike, Phoenix tends to move the other way. This is the best possible outcome for a diversifier--negative correlation means that combining two assets actually reduces portfolio variance below what either would produce alone.

The data reveals natural clusters based on how markets actually behave:

• Coastal tech/finance (New York, San Francisco): High volatility, boom-bust cycles, tightly coupled (ρ = 0.81)

• Broad growth (Los Angeles, Chicago, Seattle, Dallas, Atlanta): Lower volatility, steady growth, positively correlated with each other (ρ = 0.47 to 0.74)

• Government-driven (Washington DC): Very low volatility (1.7%), weakly correlated with everything

• Contrarian (Denver): Negatively correlated with most other markets--the strongest diversifier in this dataset

• Sun Belt recovery (Phoenix): Strongly negative with coastal markets, strongly positive with Atlanta (ρ = 0.85)

The full correlation structure is shown in the heatmap below. Blue cells indicate negative correlation (markets move in opposite directions); red cells indicate positive correlation (markets move together); white cells are near zero (independent).

Read the heatmap as a CRE manager deciding where to relocate from New York. Moving to San Francisco (ρ = 0.81) does almost nothing for diversification--the next time coastal rents spike, both markets spike together. Moving to Chicago (ρ = -0.15) or Dallas (ρ = -0.01) is dramatically better--those markets are essentially independent of New York. Moving to Phoenix (ρ = -0.45) actually reduces portfolio variance because the two markets tend to move in opposite directions.

The real question is not what size or prestige category a market belongs to, but how its rent growth correlates with the rest of your portfolio.

[1] Annual asking rent data from Newmark Research quarterly market reports (4Q25 editions), 11 markets, 2007-2025. The correlation matrix uses pairwise-maximum overlap periods (8 to 18 years depending on the pair). For the 7 markets without Newmark data (Boston, Austin, Miami, Nashville, Charlotte, Philadelphia, Portland), correlations are estimated from cluster-based priors calibrated to the empirical results.

Geographic Diversification

This section centers on a single chart, and it’s worth spending time with. The chart plots every possible geographic allocation of the portfolio on two axes: risk (standard deviation of cost per square foot) on the horizontal axis, and expected cost per square foot on the vertical axis, with lower cost at the top--following the convention used by Kritzman and other MPT practitioners where “better” is up and to the left.

Several elements appear on the chart:

Individual market dots. Each one represents a hypothetical portfolio allocated 100% to a single city. New York sits at approximately $729 per square foot with a standard deviation of $131 per square foot--high cost, high volatility. Minneapolis sits near $240 per square foot with a standard deviation of about $23 per square foot--low cost, low volatility. The other 16 markets scatter between these extremes.

The blue cloud. One thousand randomly generated geographic allocations, each a different blend of the 18 markets. These fill the feasible region--the set of all achievable cost-risk combinations. Portfolios you could construct by allocating square footage across these markets likely fall somewhere inside this cloud.

The efficient frontier. The left boundary of the cloud--a smooth curve that traces the minimum achievable risk at each cost level. This is the Markowitz efficient frontier, the same shape you’d see in an equity + bond portfolio analysis. Any portfolio on this curve is efficient: you can’t reduce risk without accepting higher cost, and you can’t reduce cost without accepting more risk.

The current allocation. Marked with an orange X at approximately $444 per square foot expected cost and $66 per square foot standard deviation. This is where the actual portfolio sits--inside the cloud, well to the right of the frontier. It is suboptimal. The same expected cost could be achieved at roughly $44 per square foot of risk instead of $66. That’s $22 per square foot of unnecessary cost volatility.

The minimum variance point. Marked with a black star near Minneapolis’s position at the bottom of the frontier--approximately $240 per square foot expected cost and $23 per square foot standard deviation. This is the theoretical floor: no allocation of any kind can produce lower risk than this. Why does Minneapolis sit at minimum variance? Three reasons compounding: it has among the lowest absolute rent levels, the lowest rent growth volatility, and the lowest correlation with high-cost coastal markets. It’s the closest thing to a risk-free asset in the occupier’s universe.

The takeaway from this chart is not that every company should move to Minneapolis, which is a nice city but too cold for some people! The takeaway is that the distance between the orange X and the efficient frontier is measurable, and that distance represents a quantifiable inefficiency in the portfolio. The geographic allocation question--which is not typically framed in these terms--is: how much of that gap can we close, and at what pace?

Before considering that question, we should examine another way of thinking about a portfolio’s risk: how do its leases relate to the broader real estate market?

Mark to Market

Geographic diversification tells you about allocation efficiency: given your total cost, are you taking more volatility than you need to? But even a geographically efficient portfolio has embedded gains and losses hiding inside individual leases. This is where mark-to-market analysis enters--not as a second optimization frontier, but as a diagnostic.

CRE portfolio managers routinely mark their leases to market. The standard exercise is straightforward: compare what you’re paying on each lease to what the market is currently asking for comparable space. If you’re paying $50/SF and the market is at $55, you’re $5 below market--embedded value. If you’re at $60, you’re $5 above--an embedded liability. This is a useful snapshot, and most portfolio reviews include some version of it.

But a snapshot is all it is. That $5 below-market position tells you where you stand today. It says nothing about where you’ll stand in three years, or five, or at lease expiration. The market will move. Your contracted rent will escalate (or not). The spread between the two will widen, narrow, or flip sign entirely--and you have no way of knowing which.

This is where the simulation adds something new. Instead of comparing contracted rent to today’s market in a single period, we simulate 1,000 possible market rent paths forward through the remaining lease term using the same GBM model that drives the geographic analysis. For each path, we compute the spread between contracted and market rent at every future year, then present-value the entire stream back to today. The result is not a single MTM number but a distribution--an expected lifecycle MTM and a spread of outcomes around it.

Think of the standard mark-to-market as a photograph. The simulation-based approach is a time-lapse. It shows not just where you stand today, but the present value of where you might stand at each point through lease expiration, across a thousand possible futures.

For a lease expiring in six months, the photograph is sufficient--the market isn’t going to move far. But for a lease with seven years remaining, the photograph is misleading. A lease that is $5 below market today could be $15 below or $8 above by year seven, depending on which rent path materializes. The lifecycle MTM captures that full range and discounts it to a single present value with confidence bands.

This has practical value in two specific situations.

First, sublease economics. If you need to vacate space before the lease expires, the lifecycle MTM tells you not just whether you’re above or below market today, but how likely that position is to hold over the sublease marketing period. A lease that looks slightly above market in a spot comparison might have a 70% probability of being well above market by the time you execute the sublease--or a 30% probability of flipping below. The distribution quantifies that uncertainty.

Second, exit prioritization. A lease with large positive lifecycle MTM--expected to remain well above market through expiration--is a liability worth acting on. If the space is underutilized, this lease should be near the top of the sublease or early-termination list. The lifecycle framing ranks all the portfolio’s embedded liabilities on the same present-value scale, accounting for how long each liability persists and how certain the estimate is.

These are diagnostic uses--they help you see the portfolio’s current state and its likely trajectory more clearly, and prioritize actions you were likely already considering. The simulation doesn’t tell you anything a seasoned CRE manager doesn’t intuit. What it does is put numbers and confidence intervals on that intuition, and it does so across the entire portfolio simultaneously rather than lease by lease.

What the MTM analysis does not do, at least in this framework, is generate a second efficient frontier that tells you how to structure deals. You might expect that comparing flat and escalating lease structures in MTM terms would reveal a preferred deal structure--and the math does show that escalating leases track market rent more closely while flat leases diverge further. This is an interesting area for further study.

Putting It Together

The geographic frontier is the primary tool. It quantifies a portfolio inefficiency that most occupiers carry but rarely measure: the cost volatility created by concentrating square footage in highly correlated markets. The MTM diagnostic complements it by identifying which specific leases represent embedded liabilities or assets--information that helps prioritize which lease events to act on first.

Consider a concrete example. Your New York leases are above market--positive MTM, an embedded liability. New York is also a high-cost, high-volatility market that pushes the portfolio to the right on the geographic frontier. If you sublease that above-market New York space and relocate to Dallas--a lower-cost, lower-correlation market--you improve your position on the frontier and shed an embedded liability in one move. The MTM diagnostic didn’t tell you to move to Dallas; the frontier did. But it told you that the New York sublease is less painful than you might have thought--or more painful--and quantified the difference.

The geographic frontier answers where. The MTM diagnostic answers which ones first. Together they turn a list of upcoming lease events into a sequenced portfolio strategy.

The Optimization Exercise

Theory is useful. But the question that matters is: can you actually move the portfolio toward the frontier using realistic lease events? Not hypothetical reshufflings of square footage, but subleases, relocations, downsizes, and renewals that a CRE team could actually execute?

We tested this with a 10-quarter trajectory--two and a half years of portfolio repositioning, one move per quarter. The moves are a realistic mix:

• Three subleases with downsizing: Sublease 55,000 SF in Midtown Manhattan, relocate a 45,000 SF operations team to Dallas (18% reduction). Sublease 45,000 SF in Century City, take 40,000 SF in Nashville. Sublease 40,000 SF in Boston’s Seaport, take 35,000 SF in Minneapolis.

• Two full relocations at lease expiry: 50,000 SF from San Francisco’s SoMa to Phoenix at half the rent. 50,000 SF from FiDi to Atlanta for the back-office team.

• Two downsize-relocations: Boston: keep a 10,000 SF executive suite, move 30,000 SF to Charlotte. Chicago: downsize from 35,000 SF to 25,000 SF in Dallas.

• Two renew-in-place with downsizing: Washington DC: renew but shrink from 35,000 to 20,000 SF (hybrid work reduction). San Francisco: renew but shrink from 40,000 to 25,000 SF (remote-first shift).

Total: 430,000 SF vacated, 360,000 SF acquired. A net reduction of 70,000 SF, reflecting the reality that many of these moves are also driven by headcount changes and space utilization trends.

The results:

The trajectory chart shows the portfolio moving from its starting position--the orange X inside the feasible region--toward the efficient frontier. An arrow traces the path from "before" to "after," a green diamond near the frontier's edge. The remaining gap between the after-point and the frontier (about $17 per square foot of risk at the post-trajectory cost level) reflects the constraints of real-world lease terms. Full optimization would require subleasing at a scale beyond practical limits. But the gap has narrowed substantially.

Every move in this trajectory is a realistic lease event. No exotic instruments. No hypothetical market conditions.

The net effect on market exposure tells the story concisely:

The portfolio shifts from a coastal-heavy concentration (high cost, high correlation, high volatility) toward a more balanced allocation across markets whose rents move more independently. The cost savings are real--$38 per square foot across the portfolio--but the risk reduction is also important, because it means the range of possible future cost outcomes has narrowed by nearly 12%.

Conclusion

When every lease event is treated as a portfolio trade, the CRE team gains a framework for why--not just what--to do next. The sublease of a Midtown lease isn’t just a cost reduction. It’s a simultaneous improvement on two dimensions of portfolio risk: geographic concentration and embedded mark-to-market exposure. The renewal of a San Francisco lease with a 15,000 SF downsize isn’t just a space utilization adjustment. It’s a deliberate reduction in the portfolio’s exposure to the most volatile rent market in the dataset.

The conversation with the CFO changes. Instead of presenting a list of lease expirations and their renewal costs, the head of real estate can say: “Our portfolio carries $66 per square foot of cost volatility against a frontier of $44. We’re repositioning 12% of the portfolio over 10 quarters to close that gap by nearly 12%, while cutting expected cost by 9%. Here’s the path, here’s the sequence, and here’s why each move matters.”

That’s a conversation grounded in the same analytical language the CFO uses for every other asset class on the balance sheet. It positions the real estate function not as a cost center executing transactions, but as a portfolio manager making deliberate allocation decisions.

The tools are the same ones Markowitz developed in 1952. They’ve been refined for equities, fixed income, commodities, and currencies. They’ve been applied to every asset class except the one that sits on the operating side of every corporate balance sheet. The math works. The data exists.

The only thing missing has been the lens. Now you have it.

---

Bibliography

Amédée-Manesme, C-O., Barthélémy, F., Prigent, J-L. “The Impact of Lease Structures on Optimal Holding Period.” Journal of Real Estate Portfolio Management 20, no. 2 (2014).

Anderson, R. I., et al. “Curvature and Contingencies: Pricing Lease Term Risk.” Journal of Real Estate Finance and Economics (2025).

Brueckner, J. K. and Rosenthal, S. S. “Tenant Riskiness and the Term Structure of Commercial Leases.” Working Paper, 2022.

Colacino, M. “Monte Carlo Simulation: Stochastic Contracting in Landlord and Tenant Lease Portfolios.” Real Estate Review (2008).

Eichenblatt, D. L. and Yeoman, J. C. “Monte Carlo Simulation: Impact of Tenant Size and Credit Quality on Office Building Risk and Return.” Real Estate Review (Fall 1999): 47ff.

Fabozzi, F. J. Fixed Income Analysis. 2nd ed. Hoboken: Wiley, 2007.

Grenadier, S. R. “An Equilibrium Analysis of Real Estate Leases.” Journal of Business 78, no. 4 (2005): 1173--1214.

Mandelbrot, B. and Hudson, R. L. The (Mis)Behavior of Markets. New York: Basic Books, 2004.

Markowitz, H. Portfolio Selection: Efficient Diversification of Investments. Malden, MA: Blackwell Publishers, 1991. Originally published 1959.

Mueller, G. R. “Refining Economic Diversification Strategies for Real Estate Portfolios.” Journal of Real Estate Research 8, no. 1 (1993): 55ff.

Mueller, G. R. and Ziering, B. A. “Real Estate Portfolio Diversification Using Economic Diversification.” Journal of Real Estate Research 7, no. 4 (1992): 375ff.

Newmark Research. U.S. Office Market Reports, Fourth Quarter 2025. Multiple metropolitan area editions. Data sourced from CoStar/Altus Data Solutions.

Tuckman, B. and Serrat, A. Fixed Income Securities: Tools for Today’s Markets. 3rd ed. Hoboken: Wiley, 2011.