Liability-Driven Investment Modelling

### The Shadow That Drives the Light: A Practical Guide to Liability-Driven Investment Modelling We spend a lot of time in our industry staring at the asset side of the balance sheet. We chase alpha, we diversify beta, we obsess over Sharpe ratios and drawdowns. It’s a seductive game, watching the curve go up and to the right. But for anyone who has sat on the trustee board of a pension fund, or managed the treasury of an insurance company, you quickly learn the harsh truth: **the asset side is just the engine; the liability side is the road.** And if the road ends abruptly, it doesn't matter how powerful your engine is. I remember sitting in a strategy meeting back in 2019, before the world went sideways. We were reviewing a client’s portfolio—a mid-sized corporate pension scheme. The funding ratio was healthy, around 105%. Everyone was smiling. Then our head of risk asked a simple question: "But what happens if rates drop 100 basis points and inflation stays sticky? Our discount rate drops, liabilities balloon, and suddenly we're at 90% funded. The assets might still be fine, but the *road* just fell off a cliff." That silence in the room was deafening. That’s when it clicked for me—and for the client—that **Liability-Driven Investment (LDI) is not just a hedging strategy; it’s the very grammar of financial survival.** It’s the discipline of admitting that the primary objective isn't to get rich; it’s to avoid going broke relative to what you owe. This article isn't an academic treatise. It’s a field guide, born from the trenches of data strategy and quantitative finance development. We're going to peel back the layers of LDI modelling, look at the messy, practical components, and understand why it's the most intellectually honest way to manage institutional money. Let’s dive in, not with the textbook definitions, but with the dirty details that actually matter. --- ### The Holy Grail: Moving Beyond the Single-Number Funding Ratio Most people think LDI is about one giant hedging trade. Buy a bunch of long-dated bonds, match your cash flows, done. If only it were that simple. The first gut-check for any modeller is realizing that a liability is not a static lump sum; it's a **a complex, multi-dimensional vector of obligations** that shifts with demographics, inflation, and longevity. The funding ratio—assets divided by liabilities—is a convenient fiction. It’s a snapshot of a moving target, and banking on that single number as your success metric is like navigating a storm using only a compass that points to where you were yesterday. In practice, the modern approach involves a granular projection. We’re not just looking at the present value of benefits; we’re building a **stochastic cash flow model** that projects every single pension payment due over the next 60 years. This isn't just an actuary's job anymore. It requires a data strategy that merges member-level data (age, salary, service years) with macro-economic scenarios. The "liability" is actually a probability distribution of cash outflows. Once you accept that, you stop trying to *match* a number and start trying to *shape* a risk profile. One of the biggest "aha" moments for our team at Originalgo happens when we show clients the concept of "LDI ratios" versus "funding ratios." Take a UK DB scheme, for instance. The funding ratio might say 95% on a low-risk discount rate. But the LDI ratio—which specifically looks at the sensitivity to interest rates and inflation—might be significantly worse. The former tells you if you have enough money on average; the latter tells you if you're going to have a heart attack when the yield curve inverts. The key insight here is that LDI modelling forces the conversation away from "are we rich enough?" to "are we safe enough?" That shift in perspective is the foundation of every other decision that follows. --- ### The Naked Driver: Deconstructing the Discount Rate If there’s a single concept that causes the most sleepless nights in a treasury department, it’s the discount rate. This is the rate used to calculate the present value of future liabilities. In LDI, the discount rate isn't just a financial metric; it's a policy decision that determines how much "risk" you are implicitly taking. If you use a high discount rate (e.g., 5%), your liabilities look smaller. You look solvent. But you're implicitly assuming your assets will earn a high return. If you use a low discount rate (e.g., a risk-free yield of 2%), your liabilities balloon, and suddenly the plan looks massively underfunded. In the world of LDI modelling, we aren't picking a random number. We are running a regression on the yield curve. The standard practice for many pension plans is to use a "gilts +" or "corporate bond yield" curve. But the modeller's dilemma is: *which* maturities matter most? If your liabilities are long-dated (30+ years), the 10-year yield is less relevant. You need to model the true term structure. I recall a project where we analyzed a client’s cash flows against the UK Gilt curve. The standard spreadsheet used a flat 1.5% discount rate. That was fine for compliance, but useless for risk management. When we applied the actual spot curve—where the 30-year yield was 1.2% and the 5-year was 2.1%—we found a massive mismatch in *sensitivity*. We use a technique called **Duration Matching**, but even that is a simplification. We build a model that calculates the PV01 (Present Value of a 01, or basis point) change for every maturity bucket. The LDI model doesn't just ask "what is the value?" It asks "**how does the value change across the curve?**" This is where the data strategy comes alive. You need yield curve data that’s not just close-of-day, but intraday. You need to model the basis risk between the discount rate you use for accounting and the hedge instruments you use for investment. If they don't move in lockstep, you have a "basis risk" black hole. We’ve seen funds lose millions not because rates moved, but because the *spread* between the discount index and the actual hedge assets widened unexpectedly. --- ### Hedging the Ghost: Inflation and the Indexation Trap Everyone remembers the 1970s. But most modern finance professionals have never traded in a truly high-inflation environment. Liability-Driven Investment isn't just about interest rates; it’s heavily skewed towards inflation. Why? Because most pension benefits are indexed, either partially or fully, to the Consumer Price Index (CPI) or RPI. This means that as inflation rises, your contractual obligations rise. But here’s the devil in the details: **the inflation you pay is not the inflation you can easily hedge.** Most LDI strategies use inflation swaps or index-linked gilts to hedge this risk. But the modelling side gets sticky. You have a "lag" issue. For example, in the UK, the RPI indexation for pensions often has an 8-month lag. So, the inflation you experience in April is based on the index from the previous August. This creates a subtle but critical modelling problem. Your hedge instrument (the swap) is based on the current index level, but your liability is tied to a lagged index. The difference is called **"inflation lag risk."** It’s a silent killer. When we build our LDI models at Originalgo, we don't just map inflation linearly. We build an autoregressive model that simulates the path of inflation, including the lag effects and possible shocks. We also have to account for the *type* of inflation. There’s headline RPI, there’s CPI, and there’s the "caps and collars" often used in pension schemes (e.g., 5% cap, 3% floor). If you model inflation as a simple 2% constant, you’re going to miss the convexity. A realistic model uses a stochastic inflation generator that does not follow a normal distribution, but rather a fat-tailed distribution to account for price shocks. I remember a specific case in 2022, when real yields went parabolic. A client of ours had a static inflation hedge that they thought was bulletproof. But because the model hadn't adequately stressed the *interaction* between nominal rates and breakeven inflation rates, their hedging ratio was off. We had to overhaul their entire data feed to capture the intraday liquidity premium in the inflation swap market. --- ### The Plan B: Dynamic Programming and the Rebalancing Dance LDI is not a "set and forget" strategy. It’s a dynamic process, a constant rebalancing act. The traditional approach is a "Level 1" hedge—static. But the sophisticated LDI model incorporates a **management strategy that reacts to the funding level**. This is where the mathematical modelling gets heavy. We’re transitioning from simple hedging to what we call "Dynamic LDI." The concept is simple: when the funding ratio is high (say 120%), you can afford to take more risk on the asset side (equities, credit). You're playing with "house money." But when the funding ratio drops (say 90%), you are essentially using borrowed capital (future contributions) and you should *de-risk*, moving assets into bonds that more closely match liabilities. The model needs to define "triggers." If funding drops by 5%, move 10% of equities to LDI pool. This is the **glide path to buy-outs**. In my experience, this is one of the hardest things to implement *operationally*. Every time the funding ratio drops, you are selling equities—often at the bottom of a market crash. It hurts. Your internal stakeholders scream "buy low, sell high!" But the LDI model ignores the market levels; it only looks at the *relative* funding ratio. I’ve had to sit in meetings and explain that the model is selling equities because the *liabilities have increased faster* than the assets, and we are taking the risk off the table to protect against further decline. It’s a behavior change. We now use **Monte Carlo simulations** that run 10,000 different market paths. We don't look at the "average" outcome. We look at the 5th percentile. The model doesn't just tell you the expected funding ratio; it tells you the probability of falling below the 75% threshold, which would trigger a "cash call" from the sponsor. That probabilistic thinking is what separates a good LDI model from a box-checking exercise. --- ### The Data Labyrinth: Collateral, Liquidity, and the Swap Curve Let’s get to the part that makes my team sweat: the plumbing. LDI is usually executed via derivatives—interest rate swaps, inflation swaps, and sometimes swaptions. Derivatives mean collateral. Collateral means liquidity. And liquidity means you need a **robust, real-time data model that tracks your counterparty exposure and margin calls**. This is where the "data strategy" aspect of my job hits the wall. Imagine you have a swap with Bank X. The swap is valuable to you (you are "in the money"). That means Bank X owes you money. But if rates move against you the next day, you owe *them* cash. In a 2020-style flash crash, the variation margin calls can be brutal—billions of dollars moving in 24 hours. A classic LDI failure isn't usually about the long-term liability structure; it’s about being wiped out by a short-term liquidity squeeze because the model didn't account for the **collateral elasticity**. I remember a case in 2022 where a client of mine had a vast LDI portfolio. The model said their surplus was safe. But they hadn't modelled the "cash drag" of posting collateral against falling Gilt prices (the infamous "Gilt crisis"). The LDI model must include a collateral module that stress-tests the portfolio against a 200bp yield rise in a short period. You have to project the Value-at-Risk of your swap portfolio *and* the cash needed to cover the margin. If you don't have that cash buffer in liquid assets (like Short-Term Investment Funds), you are forced to sell the very assets you were trying to hedge with. The ultimate paradox of LDI is that it can create systemic risk if the funding liquidity is not modelled as a primary constraint, not a secondary thought. We now run a "liquidity stress test" alongside every LDI forecast, looking at the Gap Risk (the risk that you can't raise cash fast enough). --- ### The Human Element: Building the Bridge Between Actuaries and Quants Here is something they don't teach you in CFA classes: the hardest part of LDI Modelling isn't the math; it's the *politics*. You have two distinct tribes on your hands. The Actuaries speak in terms of "best estimate assumptions," "mortality tables," and "spread calculations." They use long-term averages and are comfortable with smoothing. The Quants/Investment Bankers speak in "volatility surfaces," "day count conventions," and "regime switching." They want precision to the second decimal. As a developer, I am the bridge. I learned this the hard way. We built a sophisticated LDI model that was mathematically beautiful. It used a Heath-Jarrow-Morton framework for interest rates, which was great for the quants. But when we showed it to the actuary, she asked, "Where are my deterministic scenarios? Where is the 0.5% inflation scenario?" The quants had thrown it out because it had a low probability. The actuary shot back, "Low probability doesn't mean zero consequence. My trustee wants to see that scenario." That argument cost us two weeks. The solution wasn't more math; it was better **model governance**. We built a system that presented a "front-end" with actuarial assumptions (defined benefit cash flows) and a "back-end" with market stochastic engines. The key was to separate the *Economic Scenario Generator* (ESG) from the *Liability Calc Engine*. We allow the actuaries to have a "scenario override" function. They can force a scenario, even if it's improbable, to see the impact on the funding ratio. This creates trust. It also reveals an uncomfortable truth: LDI is not a science; it’s a decision-support tool. The model is there to clarify the trade-offs, not to mask them. In this line of work, **a model that is trustworthy is worth infinitely more than one that is mathematically perfect but opaque.** --- ### Conclusion: The Unfinished Symphony Liability-Driven Investment Modelling is, at its core, an act of humility. It admits that we cannot predict the future, but we must prepare for it. We move from a brutal mindset of "maximize return" to a mature mindset of "optimize the risk of failure." The funding ratio is the heartbeat, but the volatility of that ratio is the blood pressure. We monitor both. The journey from a simple asset allocation strategy to a fully integrated LDI framework involves technology, governance, and a cultural shift that is often painful. Where are we heading? The next frontier is **Artificial Intelligence in LDI**. We are moving away from just historical correlation to causal inference. We’re starting to use machine learning to detect regime changes in the yield curve faster than the parametric models can. We’re also looking at "Green LDI"—how to hedge liabilities while incorporating ESG factors into the collateral pool. It’s messy, but it’s necessary. The models we build today are ugly, complicated, and full of caveats. But they serve a purpose: they allow a fund manager to sleep better at night, knowing that even if the storm hits, the shadow of liability won't consume the light of the assets. --- ### A View from Originalgo Tech At ORIGINALGO TECH CO., LIMITED, we view LDI not as a standalone product but as the apex of a financial data strategy. Our work in AI finance development has taught us that the biggest failures in this space come from data silos—when the market data team doesn't talk to the liability data team. Our perspective is this: **the future of LDI lies in the "live" model.** This means ingesting real-time cash flows, real-time collateral positions, and real-time yield curves. We are building platforms where the LDI model is not run monthly, but continuously, using streaming data. This allows treasury teams to see their exact hedge ratio, their exact liquidity headroom, and—most importantly—their *forecasted* funding ratio volatility at any given minute. We believe that the administrative drudgery of collateral management (which is our specialty) can be automated and optimized with algorithmic decision support. We are not just building a model; we are building the nervous system for the institutional balance sheet. It’s not about the destination; it’s about the velocity of the journey—and we ensure that velocity is under your control.