User-provided solar-panel wallpaper
UNIVERSITY OF BOLOGNA
PhD in Statistical Sciences
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
From local uncertainty to electricity-linked
and spatial hedging
CANDIDATE
Beniamino Sartini
SUPERVISOR
Silvia Romagnoli
28 September 2026
2
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
THESIS ROADMAP
From solar variability to contracts and hedging decisions
CONTEXT
PV output depends on variable solar radiation
→
FINANCIAL PROBLEM
Production shortfalls expose the producer to revenue losses
→
PROPOSED CONTRACTS
GHI puts transfer part of this risk using CAMS settlement
MODELS AND APPLICATIONS
Local bounded radiation
Local bounded radiation
Eight European cities
From the GHI law to a solar power
producer with POA exposure.
Bounded GHI → European evidence → POA hedge
Radiation and electricity
Radiation and electricity
Joint price-volume exposure
Continuous-time representation,
price-linked contracts and futures.
Joint exposure → valuation → futures hedge
Spatial dependence
Spatial dependence
Portfolios across locations
Joint radiation risk and contracts
at reference locations.
Portfolio VaR → reference-site cross-hedge
Three coordinated applications: plant technology, radiation-electricity exposure and geographical basis risk.
3
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
MOTIVATION - FROM WEATHER TO FINANCIAL RISK
Cloud variability becomes revenue risk
Daily global horizontal irradiation (GHI): solar energy received by a horizontal surface, in kWh/m².
Observed daily GHI and fitted clear-sky reference at eight European locations
Estimated clear-sky radiation and realised daily GHI at the eight European locations in 2024; radiation in kWh per square metre.
Sun and cloud
PHYSICAL SIGNAL
Observed GHI falls below its clear-sky reference.
Coins
FINANCIAL CONSEQUENCE
Lower production becomes a revenue loss through plant geometry and electricity prices.
Sprout
A possible solution
A GHI put pays when radiation falls below the contractual strike.
4
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
THESIS LOGIC
The exposure determines the information the model must deliver
Common solar-radiation risk
Solar-radiation risk
Plant technology
GHI versus POA
Marginal
distribution
Bilateral hedge
Joint price-volume risk
Radiation and power prices
Joint temporal
moments
Derivative plus futures
Geography
Several sites
Cross-location
moments
Reference-site cross-hedge
5
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
MODEL ARCHITECTURE
From observed radiation to a predictive GHI distribution
RADIATION SCALE
TRANSFORMED SCALE
RADIATION SCALE
1
Clear-sky
reference
Express GHI as a shortfall from the fitted clear-sky reference.
2
Bounded
link
Map the normalised shortfall to an unbounded state.
3
Unbounded domain
Represent daily memory and changes in the conditional scale.
4
Monthly mixture
Two Gaussian components describe the innovation shape in each month.
5
Inverse
transformation
Return the predictive law to GHI, on its moving bounded support.
One step ahead forecast: an exact mixture on the transformed scale, mapped back to GHI.
6
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
BOUNDED RADIATION
The transformation preserves a moving radiation support
GHI and clear-sky radiation above; transformed radiation below, Bologna and Oslo.
FROM GHI DOMAIN TO $(0,1)$
$$ X_t^{R}=1-\frac{R_t}{C_t},\qquad X_t^{R,\prime}=\frac{X_t^{R}-\alpha^{\mathrm{tr}}}{\beta^{\mathrm{tr}}} $$
$$ Y_t=g_{\mathrm{link}}(X_t^{R,\prime})\in\mathbb{R} $$
The link from $(0,1) \to \mathbb{R}$ is a monotone bijection.
RETURN TO THE RADIATION SCALE
$$ C_t(1-\alpha^{\mathrm{tr}}-\beta^{\mathrm{tr}})<R_t<C_t(1-\alpha^{\mathrm{tr}}) $$
GHI with the clear-sky envelope and transformed radiation Y for Bologna and Oslo over 2015-2024; the divider marks 1 January 2023.
solar risk driver
normalized shortfall
$C_t$ is the fitted clear-sky reference
$\alpha^{tr}$ and $\beta^{tr}$ normalise the shortfall to $(0,1)$
FROM $(0,1)$ TO AN UNBOUNDED DOMAIN
7
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
LOCAL MODEL: MATHEMATICAL SPECIFICATION
Seasonality, daily memory and monthly innovation mixtures
BOUNDED TRANSFORMATION
$$ Y_t=\log\!\left[-\log\!\left(\frac{1-R_t/C_t-\alpha^{\mathrm{tr}}}{\beta^{\mathrm{tr}}}\right)\right] $$
SEASONAL VARIANCE
$$ \bar{\sigma}_t^2=b_0+b_1\cos\!\left(\frac{2\pi t}{365}\right)+b_2\sin\!\left(\frac{2\pi t}{365}\right) $$
SEASONAL MEAN
$$ \bar{Y}_t=a_0+a_1\cos\!\left(\frac{2\pi t}{365}\right)+a_2\sin\!\left(\frac{2\pi t}{365}\right) $$
GARCH DYNAMICS
$$ \sigma_t^2=\omega+\sum_{k=1}^{r}\alpha_k\tilde{\varepsilon}_{t-k}^{2}+\sum_{k=1}^{s}\beta_k\sigma_{t-k}^2, \quad \tilde{\varepsilon}_t=\varepsilon_t/\bar{\sigma}_t $$
ARMA DYNAMICS
$$ \tilde{Y}_t=\sum_{k=1}^{p}\phi_k^{\mathrm{AR}}\tilde{Y}_{t-k}+\sum_{k=1}^{q}\theta_k^{\mathrm{MA}}\varepsilon_{t-k}+\varepsilon_t, \quad \tilde{Y}_t=Y_t-\bar{Y}_t $$
MONTHLY GAUSSIAN MIXTURE
$$ u_t\mid B_t=b\sim\mathcal{N}\!\left(\mu_t^{(b)},\sigma_t^{2(b)}\right),\quad b\in\{0,1\} $$
1
2
3
4
5
6
8
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
LOCAL MODEL VALIDATION
Eight European cities test the local model out of sample
Oslo, Amsterdam, Berlin, Paris, Milan, Bologna, Rome and Palermo
COMMON DATA SOURCE
Daily GHI, 2005-2024
TRAINING 
2005-2022
2023-2024
Oslo, Amsterdam, Berlin, Paris, Milan, Bologna, Rome and Palermo. Local models are fitted separately.
kWh/m² per daily observation from CAMS data.
TEST 
WHAT IS CHECKED
Residual dependence and calibration: CDF at observations (PIT), error metrics on out-of-sample forecasts and test on the violations of the VaR. 
PIT uniformity is not rejected at 5% for the complete models; selected city-month and tail checks still show departures.
9
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
PROPOSED STANDARDISED CONTRACT
One GHI reference, one transparent settlement rule
The GHI put pays the monetised shortfall below strike K.
TERMS DECLARED AT INCEPTION
WHERE
Reference coordinate $\ell$
WHEN
$$ \Gamma_{\ell}^{\mathrm{put}}(T,K;R)=\zeta^{R}\left(K-R_{\ell,T}\right)^{+} $$
Maturity $T$ or delivery period
PROTECTION
Strike $K$ and monetary tick $\zeta^R$
SETTLEMENT DATA
CAMS
Copernicus Atmosphere Monitoring Service
Solar Radiation Time Series: realised daily GHI
The common index standardises settlement. Plant-specific POA determines the remaining basis risk.
Solar put payoff as a function of realised daily GHI. Below K, each unit of shortfall pays the monetary tick ζᴿ.
10
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
HEDGING MECHANISM
The derivative offsets the low-radiation tail
Physical exposure, solar put payoff and hedged P&L across realised GHI
$$ \mathrm{P\&L}_{\mathrm{hedged}}=\mathrm{exposure}+q\, (\mathrm{payoff}-\mathrm{price}) $$
Low radiation
Physical P&L falls
Solar put
Contract payoff rises
Hedge quantity q
Sets the strength of the offset
The hedge transfers part of the low-radiation loss without changing the physical exposure.
Schematic P&L against realised GHI: physical exposure, put payoff and the combined hedged position.
11
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
PREDICTIVE DISTRIBUTION
Conditional GHI densities above and CDFs below at four forecast horizons in Bologna
PDF | upper panels
The density describes how probability is distributed across possible GHI values.
CDF | lower panels
The curve gives the probability that GHI is below a chosen threshold.
How this relates to a solar put
A put pays when GHI falls below the strike. The CDF at the strike gives the probability of payment.
The expected payoff also depends on the size of the shortfall:
Model approximation
Monte Carlo
Historical
The GHI distribution determines exercise risk and expected payoff
$$ F_R(K;t,T)=\mathbb{P}(R_T\leq K\mid\mathcal{F}_t) =\mathbb{P}(\Gamma_T>0\mid\mathcal{F}_t) $$
PROBABILITY TO EXERCISE THE PUT
EXPECTED PAYOFF
$$ \mathbb{E}\{\Gamma_T\mid\mathcal{F}_t\}=\zeta^R\int_{-\infty}^{K}(K-r)f_R(r;t,T)\,dr $$
Bologna | Forecast origin: 1 January 2024 | Four forecast dates
12
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
TECHNOLOGICAL BASIS RISK
A GHI hedge remains exposed to plant geometry
Incoming solar radiation received by a horizontal GHI settlement panel and a tilted POA exposure panel
-100%
-50%
0
Full-GHI exposure across cities
about -96% to -90%
Amsterdam, fixed annual tilt
about -38%
Amsterdam, seasonally adjusted tilt
about -26%
THE GHI-POA LINK
$$ R_{t_m}^{\text{spp}(b)} = \text{BNI}_{t_m} \cos(\text{AOI}_{t_m}^{\text{spp}(b)}) + \text{DHI}_{t_m} \left(\frac{1- \cos(\theta_{\text{tilt}}^{\text{spp}(b)})}{2}\right) + \text{alb}^{\text{spp}(b)} \cdot \text{GHI}_{t_m} \left(\frac{1+ \cos(\theta_{\text{tilt}}^{\text{spp}(b)})}{2}\right) $$
Correlation of aggregated annual POA with the annual solar put-index payoff.
ANNUAL POA / PUT-INDEX CORRELATION
Beam Normal Irradiance
Direct Horizontal Irradiance
13
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
LOCAL APPLICATION: PHOTOVOLTAIC PRODUCER
The same GHI contract hedges different PV configurations
Daily PV cash flows: Palermo, Bologna and Oslo, under full GHI, annual fixed tilt, east-west and seasonal tilt.
ONE SETTLEMENT INDEX
GHI is common to every configuration. POA changes with tilt and orientation.
FOUR CONFIGURATIONS
Full-GHI benchmark
South-facing, annual fixed tilt
East-west
South-facing, seasonal tilt
() × Unhedged ● Hedged (option pays; zero payoff) 
Daily photovoltaic cash flows in EUR for Palermo, Bologna and Oslo (rows), under full-GHI, fixed annual tilt, east-west and seasonally adjusted configurations (columns).
The hedge supports low-output cash flows. Its effectiveness depends on how closely POA tracks the GHI index.
14
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
FROM PRODUCTION RISK TO JOINT MARKET RISK
Revenue depends jointly on radiation and electricity prices
PHYSICAL PRODUCTION
Radiation → POA → output
Plant configuration determines how radiation becomes electricity production.
MARKET REVENUE
Output × electricity price
A low-output day has a different financial effect when the electricity price changes.
→
We now model radiation and electricity prices to value price-weighted protection and an electricity-futures hedge.
A CONTINUOUS-TIME REPRESENTATION
15
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
LOG-ELECTRICITY PRICE
$$ X_t^{E}=\log(E_t) $$
$$ dX_t^{E}=\kappa^E\!\left(\mu_X-\frac{\sigma_X^2}{2\kappa^E}-X_t^{E}\right)dt+\sigma_X\,dW_t $$
$\kappa^E > 0$: mean-reversion speed
$\sigma_x$ > 0: diffusion coefficient
$\mu_x$: drift
FROM PRODUCTION RISK TO JOINT MARKET RISK
Revenue depends jointly on radiation and electricity prices
$$ dY_t=d\bar{Y}_t+\theta(\bar{Y}_t-Y_t)\,dt+\mu_Y(t,B_t)\,dt+\sigma_Y(t,B_t)\,dM_t $$
SOLAR RADIATION
$$ Y_t=g_{\mathrm{link}}(X_t^{R,\prime}) $$
$$ dM_t=B_t\,dW_{1,t}+(1-B_t)\,dW_{0,t} $$
$$ \sigma_Y(t,b)=\bar{\sigma}_t\sigma_{\mathrm{m}(t)}^{(b)} $$
$$ \mu_Y(t,b)=\bar{\sigma}_t\mu_{\mathrm{m}(t)}^{(b)} $$
$\theta > 0$ mean reversion speed
$\bar{Y}_t$ is the deterministic annual level
$B_t \in \{1,0\}$ is a continuous-time Markov chain.
16
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
APPROXIMATION
A two-Gaussian reduction makes repeated valuation tractable
COMPLETE PATH MIXTURE
Regime sequences grouped by terminal CTMC state zero or one
GROUP PATHS BY TERMINAL STATE
Match the exact bridge mean and variance within each group
›
TWO CONDITIONAL GAUSSIAN COMPONENTS
Each terminal-state path mixture becomes one Gaussian
TERMINAL STATE 1
Moment-matched Gaussian component conditional on terminal state zero
TERMINAL STATE 0
Moment-matched Gaussian component conditional on terminal state one
›
›
exact conditional identities
path-dependent exact law
two-component moment-matched approximation
Different regime paths can end in the same state. Conditional on the full path, the transformed law is Gaussian. Collapsing these paths makes payoff calculations tractable.
17
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
CONTRACT DESIGN · BOLOGNA, ROME AND PALERMO
Solar contracts and futures lower daily P&L variance
SoRad: radiation put
$$ \Gamma^{\mathrm{SoRad}}=\zeta^{R}\left(K-R_T\right)^{+} $$
Targets volumetric risk
SoREd: price-weighted radiation put
$$ \Gamma^{\mathrm{SoREd}}=\zeta^{E}E_T\left(K-R_T\right)^{+} $$
Targets price-weighted volume risk
x
−25.5% to −19.6%
−40%
−20%
0%
Without futures
−16.0% to −11.6%
With futures
−19.2% to −16.3%
−40%
−20%
0%
SoRad
Radiation shortfall
Electricity price
Radiation shortfall
Change in average realised daily P&L variance, 2014–2023. Negative values mean lower variance.
Changes are relative to each contract’s unhedged exposure; ranges span the three cities. −25% means 25% lower variance.
18
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
RADIATION-ELECTRICITY APPLICATION
Daily P&L and variance change, with and without futures
Bologna, Rome, Palermo: daily P&L for SoREd without futures (top row) and SoREd plus futures (bottom row).
2018 daily P&L; columns: Bologna, Rome, Palermo. Rows: SoREd, SoREd + futures, SoREdIDX + futures (annual index).
Contract / hedge
Bologna
Rome
Palermo
SoRad
−25.5%
−19.6%
−20.1%
SoREd without futures
−16.0%
−11.7%
−11.6%
SoREd with futures
−19.2%
−16.3%
−16.5%
VARIANCE CHANGE
100 × (average hedged variance /
average unhedged variance − 1), 2014–2023.
Each contract uses its corresponding unhedged exposure.
−25% means that hedging lowers variance by 25%.
revised
19
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
FROM A PLANT TO A PORTFOLIO
Several locations add a new source of risk
DISTRIBUTED PRODUCTION
Portfolio risk depends on which locations experience shortfalls on the same dates.
Joint radiation law → portfolio VaR
+
STANDARDISED REFERENCE CONTRACTS
The contract may settle at a reference location while physical exposure lies elsewhere.
Cross-location covariance → cross-hedge
The spatial model supplies the joint information needed for both decisions.
What happens when low radiation affects several sites together?
20
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
A regional grid supports prediction away from estimation sites
Emilia-Romagna estimation grid and held-out city locations.
540 ESTIMATION SITES
Daily GHI over 2005-2022
Parameters fitted across the regional grid
8 EXCLUDED LOCATIONS
Local parameters and histories must be reconstructed before evaluating forecasts at these sites.
The same spatial information is then used for portfolio risk and reference-site hedging.
Emilia-Romagna grid of 540 estimation sites and eight excluded validation locations; the additional Forlì-Cesena point provides geographic context.
SPATIAL MODEL: DATA
21
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
SPATIAL DEPENDENCE AND THE PREDICTIVE LAW
Two dependence layers shape the bivariate radiation density
TWO LOCATIONS, FOUR JOINT STATES
$$ f_{\mathbf{Y}}(\mathbf{y};t,t+1)=\sum_{\mathbf{b}\in\{0,1\}^{2}}\pi_{\mathbf{B}_{t+1}}^{(\mathbf{b})}\,\phi_2\!\left(\mathbf{y};\mathbf{M}_{\mathbf{Y}}^{(\mathbf{b})},\boldsymbol{\Sigma}_{\mathbf{Y}}^{(\mathbf{b})}\right) $$
JOINT RADIATION DENSITY AT TWO SITES
Joint radiation density: surface and contours for locations 1 and 532.
RETURN TO BOUNDED GHI
$$ f_{\mathbf{R}}(\mathbf{r};t,T)=f_{\mathbf{Y}}\!\left(\mathbf{y}_{T}(\mathbf{r});t,T\right)\prod_{\ell=1}^{2}\frac{\left|g_{\mathrm{link}}^{\prime}(\eta_{\ell,T}(r_{\ell}))\right|}{\beta_{\ell}^{\mathrm{tr}}C_{\ell,T}} $$
One-step joint radiation density for grid locations 1 and 532: density surface and contours. Each axis represents daily GHI at one site.
LAYER 1 - LATENT STATES
Adverse states can occur together across locations.
LAYER 2 -  LATENT INNOVATIONS
Gaussian-shock dependence changes the within-state covariance $\Sigma_Z$.
22
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
PORTFOLIO RISK: EMPIRICAL COMPARISON
Spatial dependence changes the portfolio downside threshold
Five-location realised P&L and 5 percent lower-tail VaR under spatial dependence and independence.
Realised unhedged P&L in 2023 (orange), 5% lower-tail VaR (red), and violations (blue triangles), under spatial dependence and independence. Five locations | Bologna, Parma, Ravenna, Modena and Reggio Emilia | 2023
VIOLATIONS OF THE VALUE AT RISK (5%)
6.8%
19.7%
Spatial dependence
Independence




Adverse states can occur jointly
Joint shortfalls are suppressed by construction
A monetary exposure to $d$ locations: for the $a$-agent the weights $\mathbf{w}_a$ convert each site’s GHI into monetary exposure.
CONDITIONAL DISTRIBUTION
$$ F_{\Pi_a}(p;t,T)=\Pr(\Pi_{a,T}\le p\mid\mathcal F_t)=\int_{\mathbb R^d}\!\mathbf 1_{\{\mathbf w_a^\top\mathbf r_T(\mathbf y)\le p\}} \times f_{\mathbf Y}(\mathbf y;t,T)\,\mathrm d\mathbf y $$
$$ \Pi_{a,T}=\mathbf w_a^\top\mathbf R_T $$
The joint latent density is integrated over all site configurations whose portfolio value is at most p.
$$ \operatorname{VaR}_{0.05\mid t}=\inf\{p:F_{\Pi_a}(p;t,T)\ge0.05\} $$
23
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
REFERENCE-SITE CROSS-HEDGING
Contracts can hedge exposure at a different location
Spatial market map with eight exposure locations and 36 reference locations.
PHYSICAL EXPOSURE
Eight exposure locations
Three producers with physical portfolios
TRADABLE CONTRACTS
36 reference locations
Three additional financial agents
Cross-location exposure-payoff covariances determine how much protection each reference contract provides.
Estimation grid, eight exposure locations, 36 tradable reference locations, and the nearest-reference set used in restricted cross-hedging.
1
2
3
CROSS-LOCATIONS
Prices and hedge quantities 
depend on cross-location covariances
24
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
FROM SPATIAL MOMENTS TO CONTRACT DEMAND
Individual hedges balance expected P&L and its variance
EXPOSURE AND HEDGED P&L AT MATURITY
$$ \Pi_{a,T}=\mathbf w_a^\top\mathbf R_{a,T},\qquad\widetilde\Pi_{a,T}=\Pi_{a,T}+\mathbf q_a^\top(\boldsymbol\Gamma_T-A\mathbf V) $$
MEAN–VARIANCE OBJECTIVE
$$ \max_{\mathbf q_a}\;\mathbb E_t\{\widetilde\Pi_{a,T}\}-\frac{\nu_a}{2}\mathbb{V}_t\{\widetilde\Pi_{a,T}\},\qquad\nu_a>0 $$
$$ \mathbf q_a(\mathbf V)=\boldsymbol\Sigma_{\Gamma}^{-1}\!\left[\frac{\mathbf M_{\Gamma}-A\mathbf V}{\nu_a}-\mathbf c_a\right] $$
OPTIMAL QUANTITY FOR THE $a$-agent as function of the price $\mathbf{V}$
AGGREGATE INPUTS AND CLEARING
$$ \mathcal R_\nu=\sum_{a=1}^{N_A}\nu_a^{-1},\qquad\mathbf c=\sum_{a=1}^{N_A}\mathbf c_a,\qquad\sum_{a=1}^{N_A}\mathbf q_a(\mathbf V)=\mathbf q_t^{\mathrm{ext}} $$
$N_A$: number of agents; 
$\mathcal{R}_ν$: total risk tolerance; 
$c$: aggregate payoff-exposure covariance;
$\mathbf{q}_t^{ext}$: net supply to the agents.
$$ \begin{aligned} {} & \mathbf{M}_{\Gamma}=\mathbb{E}_t\{\boldsymbol\Gamma_T\}, \\ & \boldsymbol\Sigma_{\Gamma}=\mathbb{C}v_t\{\boldsymbol\Gamma_T,\boldsymbol\Gamma_T\} \\ & c_a=\mathbb{C}v_t\{\boldsymbol\Gamma_T,\Pi_{a,T}\} \end{aligned} $$
EQUILIBRIUM PRICE
$$ \mathbf V^{\mathrm{mv}}=D\!\left[\mathbf M_{\Gamma}-\frac{\mathbf c+\boldsymbol\Sigma_{\Gamma}\mathbf q_t^{\mathrm{ext}}}{\mathcal R_\nu}\right] $$
$D = A^{-1}$ is the discount factor. With only internal trading, $\mathbf q_t^{\mathrm{ext}} = 0$ and the equilibrium quantities sum to zero.
1
3
2
4
5
25
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
A PRODUCER WITH A PORTFOLIO
The portfolio hedge reduces variance and supports low-output days
Producer 1 | Seven exposure locations | Combined market: contracts at exposure and reference sites
DAILY P&L IN 2023
Producer 1 daily P&L in 2023, unhedged versus hedged.
Daily P&L in 2023 (EUR, by day of year) × Unhedged; coloured points: hedged. Black curve: seasonal benchmark; blue: clear-sky.
DISTRIBUTION OF ANNUAL P&L
Kernel densities of annual P&L in EUR over 2005-2023.
Annual P&L over 2005-2023, EUR. Rescaled kernel density estimates: black unhedged, red hedged.
-28%
+29%
TCM: mean P&L on dates when the unhedged position falls below its seasonal benchmark.
DAILY STATISTICS, 2005-2023
P&L variance
Target-conditioned mean
DAILY STATISTIC
UNHEDGED
HEDGED
Variance (EUR²)
11,169
8,091
TCM (EUR)
120
155
26
B. Sartini
Modelling Solar Radiation Risk: Bounded Dynamics, Spatial Dependence and Hedging
28 September 2026
CONCLUSION
Solar-radiation derivatives make volume risk contractible
The payoff is triggered by an observable GHI shortfall.
Its value and hedge ratio depend on plant technology, electricity prices and location.
The model should deliver the conditional quantity required by that decision.
Thank you
Beniamino Sartini
Mathematical Finance
Solar Energy Risks:
Stochastic Radiation Modeling
and Optimal Hedging Strategies
S. Romagnoli & B. Sartini
10.1111/mafi.70054
Scan to open DOI 10.1111/mafi.70054
Applied Energy
A fuzzy-and-fair framework for solar irradiance modeling and derivative pricing: Bridging photovoltaic production risk and climate-linked finance
S. Romagnoli & B. Sartini
10.1016/j.apenergy.2025.127139
Scan to open DOI 10.1016/j.apenergy.2025.127139