diff --git a/src/models/GenericHybridModel.jl b/src/models/GenericHybridModel.jl
index 2587d208..8037fd52 100644
--- a/src/models/GenericHybridModel.jl
+++ b/src/models/GenericHybridModel.jl
@@ -103,7 +103,7 @@ function constructHybridModel(
     @assert all(n in all_names for n in neural_param_names) "neural_param_names ⊆ param_names"
 
     # if empty predictors do not construct NN
-    if length(predictors) > 0
+    if length(predictors) > 0 && length(neural_param_names) > 0
 
         in_dim = length(predictors)
         out_dim = length(neural_param_names)
diff --git a/src/utils/loss_fn.jl b/src/utils/loss_fn.jl
index 786a8d5c..f369baa2 100644
--- a/src/utils/loss_fn.jl
+++ b/src/utils/loss_fn.jl
@@ -127,6 +127,29 @@ function loss_fn(ŷ, y, y_nan, ::Val{:kgeLoss})
     )
 end
 
+function loss_fn(ŷ, y, y_nan, ::Val{:β})
+    ŷv = ŷ[y_nan]
+    yv = y[y_nan]
+
+    μ_s = mean(ŷv)
+    μ_o = mean(yv)
+
+    β = μ_s / μ_o
+
+    return β
+end
+
+function loss_fn(ŷ, y, y_nan, ::Val{:α})
+    ŷv = ŷ[y_nan]
+    yv = y[y_nan]
+
+    σ_s = std(ŷv)
+    σ_o = std(yv)
+    α = σ_s / σ_o
+
+    return α
+end
+
 # Kling–Gupta Efficiency metric (to MAXIMIZE, e.g. for reporting)
 function loss_fn(ŷ, y, y_nan, ::Val{:kge})
     kge_loss = loss_fn(ŷ, y, y_nan, Val(:kgeLoss))
diff --git a/test/test_loss_fn.jl b/test/test_loss_fn.jl
index 03e57453..519f86e1 100644
--- a/test/test_loss_fn.jl
+++ b/test/test_loss_fn.jl
@@ -55,6 +55,12 @@ using EasyHybrid: bestdirection, isbetter, check_training_loss, Minimize, Maximi
         # KGE test (1 - KGE Loss)
         @test loss_fn(ŷ, y, y_nan, Val(:kge)) ≈ 1.0 - kge_loss
 
+        # β test (mean ratio)
+        @test loss_fn(ŷ, y, y_nan, Val(:β)) ≈ β
+
+        # α test (standard deviation ratio)
+        @test loss_fn(ŷ, y, y_nan, Val(:α)) ≈ α
+
         # PBKGE Loss test (Partial Kling-Gupta Efficiency)
         μ_s = mean(ŷ)
         μ_o = mean(y)
@@ -123,6 +129,15 @@ using EasyHybrid: bestdirection, isbetter, check_training_loss, Minimize, Maximi
         @test loss_fn(ŷ, y, y_nan, Val(:pbkgeLoss)) ≈ pbkge_loss
         @test loss_fn(ŷ, y, y_nan, Val(:pbkge)) ≈ 1.0 - pbkge_loss
 
+        # β test with NaN handling
+        @test loss_fn(ŷ, y, y_nan, Val(:β)) ≈ β
+
+        # α test with NaN handling
+        σ_s = std(valid_ŷ)
+        σ_o = std(valid_y)
+        α = σ_s / σ_o
+        @test loss_fn(ŷ, y, y_nan, Val(:α)) ≈ α
+
         # Test NaN handling for generic functions
         @test loss_fn(ŷ, y, y_nan, simple_loss) ≈ mean(abs2, valid_ŷ .- valid_y)
         @test loss_fn(ŷ, y, y_nan, (weighted_loss, (2.0,))) ≈ 2.0 * mean(abs2, valid_ŷ .- valid_y)